Platform atlas
What this is
Platform atlas - a self-documenting view of what the SPICE platform is doing right now. Every cycle that ships new parts (ContentTypes, Workflows, Tools, Listeners) auto-appears here because the projections below read the live manifest, not a hand-maintained registry.
The current horizon is dogfooded - the platform plans its
own next moves through V12.8 Tp Projects + Tasks rows. Click
into /sites/Projects/Lists/Projects/Item/V14-HORIZON
(or V13-HORIZON for the closed previous arc) to
see the horizon and advance task status via V12.2 FormView
"Next step" buttons. The same row is reachable via MCP
tool_advance_list_item for agent progress.
V14 horizon (open) - BaaN Tm Manufacturing + worked-example generalizations
- Strategic frame: Tm is the next BaaN package (after Tc/Td/Tf/Ti/Tp). The five Tm primitives (BOM / Routing / WorkCenter / ProductionOrder / MRP) generalize past discrete manufacturing into content production, software releases, sprint planning, surgical scheduling, ...
- Receipts queued: V14.1 ships Tm primitives (manifest-only); V14.2 wires the 5th cross-module listener (Tm -> Ti consumption/receipt); V14.3 deterministic MRP-lite; V14.4-V14.5 ship N>=2 worked-example portals on the same ContentTypes (zero new C#); V14.6 capacity+load card; V14.7 parent-ref Lookup sweep; V14.8 horizon debrief.
- Carry-over from V13.11.4: V13.3 A2A/ACP, V13.4 real-LLM smoke (now unblocked-on-payload by V13.6 line items), kanban POST /advance.
V12 horizon snapshot (closed - banked in Section 13.10)
- Foundation (V12.0a-V12.1): ISO 4217/3166/639 + UN-CEFACT + GAAP termsets; 6 master-data ContentTypes (BusinessPartner / Item / Address / BankAccount / ExchangeRate / ChartOfAccount).
- Workflow seam (V12.0b-V12.3): Transitions block + FormView Next-step + MCP advance. Channel symmetry on the new surface (HTTP + MCP).
- 4 operational modules (V12.4 / V12.6 / V12.7 / V12.8): Td Distribution + Tf Finance (UBL-aligned) + Ti Inventory + Tp Project. All Core-only csproj, all pure manifest.
- 3 cross-module wires (V12.7b / V12.7c): Td PO.Received -> Ti StockMove auto-create; Td SO.Shipped -> Tf Invoice auto-create; Tf Invoice.Paid -> Tf JournalEntry auto-create. N=3 receipt for the IItemAdvancedListener pattern.
- Hub projections (V12.9 ErpPortfolio + this Atlas): read-across cross-site views; hubs never write per-dept data.
Drill-down endpoints
/about- live platform snapshot (Plugins / Apps / Authentication / Cost)./admin- operator dashboard with service catalogue + Hub primitives./agency/graphs- V11.19 knowledge-graph read endpoint./mcp/jsonrpc- agent surface (V11.25/V11.26 bridges).spice://parts/KB.V12HorizonDebrief- the V12 horizon's banked wisdom (MCP resource).
Horizon projects (V13 + V14)
| Source | ID | Title | CreatedAt | Author | ProjectCode | ClientSku | StartDate | EndDate | CurrencyCode | TotalBudget | Status | Project type | Notes |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Projects | app-pycoral-clone | pycoral (clone) | 2026-10-01T10:02:51.9625253+00:00 | claude-code | APP-PYCORAL-CLONE | platform-self | 2022-03-04T00:00:00Z | Euro | 0 | Closed | App | ||
| Projects | app-jarvis-clone | Jarvis (sukeesh clone) | 2026-10-01T10:02:51.9503080+00:00 | claude-code | APP-JARVIS-CLONE | platform-self | 2024-01-24T00:00:00Z | Euro | 0 | Closed | App | ||
| Projects | app-cvmanager | cvmanager | 2026-10-01T10:02:51.9375680+00:00 | claude-code | APP-CVMANAGER | platform-self | 2026-10-01T00:00:00Z | Euro | 0 | Closed | App | ||
| Projects | app-stock-risk | stock-risk-analyzer | 2026-10-01T10:02:51.9204585+00:00 | claude-code | APP-STOCK-RISK | platform-self | 2026-10-01T00:00:00Z | Euro | 0 | Closed | App | ||
| Projects | app-genesis-farm | genesis-farm | 2026-10-01T10:02:51.9067299+00:00 | claude-code | APP-GENESIS-FARM | platform-self | 2026-10-01T00:00:00Z | Euro | 0 | Closed | App | ||
| Projects | app-vibevoice | vibevoice | 2026-10-01T10:02:51.8917033+00:00 | claude-code | APP-VIBEVOICE | platform-self | 2026-08-11T00:00:00Z | Euro | 0 | Active | App | ||
| Projects | app-docker-stacks | docker-stacks | 2026-10-01T10:02:51.8684185+00:00 | claude-code | APP-DOCKER-STACKS | platform-self | 2026-09-22T00:00:00Z | Euro | 0 | Active | App | ||
| Projects | app-n8n-workflows | n8n-workflows | 2026-10-01T10:02:51.8144004+00:00 | claude-code | APP-N8N-WORKFLOWS | platform-self | 2026-10-01T00:00:00Z | Euro | 0 | Active | App | ||
| Projects | app-aw-skills | angelsworks-skills | 2026-10-01T10:02:51.7878131+00:00 | claude-code | APP-AW-SKILLS | platform-self | 2026-09-22T00:00:00Z | Euro | 0 | Active | App | ||
| Projects | app-ooda-plugin | ooda-plugin | 2026-10-01T10:02:51.0933567+00:00 | claude-code | APP-OODA-PLUGIN | platform-self | 2026-06-25T00:00:00Z | Euro | 0 | Active | App |
Showing 10 of 34. Across 1 source site.
Backlog tasks (V13 + V14)
| Source | ID | Title | CreatedAt | Author | TaskNumber | ProjectCode | TaskTitle | AssigneeSku | DueDate | EstimateHours | Status | Body | AssignedTo | Parent task | PercentComplete | Comments | Predecessors | Related Items |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Projects | pmo-4 | PMO.4 - The pmo seat runs unattended - watch its first week | 2026-10-03T20:58:10.4911466+00:00 | claude-code | PMO.4 | PMO | pmo seat runs unattended | 2026-10-10 | 2 | In Progress | Since 2026-10-03 23:00 the pmo seat runs on this PC from the Startup folder (SPICE-PMO.cmd -> tools/spice-worker/run-seat.ps1), Ollama qwen3-coder on .69, polling every 5 minutes; log %TEMP%/spice-worker-pmo.log. Give it work by assigning a ProjectTask to pmo. Watch: Deferred tasks (review them), partial jobs (PMO.3 commented 4 of its overbooked days), TPL.35 parked Waiting (too big for this runtime). Later: a Windows service or the .69 sandbox (needs admin / the steward); DeepSeek for heavier jobs. | claude-code (seat) | (claude-code, 2026-10-03 22:23 UTC) (claude-code, 2026-10-04) Scoped: pmo writes on Projects only (proven as the seat), read elsewhere, tools allow-listed. Model confirmed by the operator: qwen3-coder:30b-aw on Ollama .69. | |||||
| Projects | pmo-3 | PMO.3 - Plan check: overbooked days for claude-code, the next two weeks | 2026-10-03T20:25:02.7562884+00:00 | claude-code | PMO.3 | PMO | Plan check: overbooked days | 2026-10-04 | 1 | Completed | Call query_assigned_work once with owner=claude-code, from=2026-10-05, to=2026-10-16. Add up EstimateHours per DueDate (the date part). For every day above 6 hours, call tool_update_list_item once on the LARGEST task of that day: site=Projects, list=ProjectTasks, itemId=<its key>, field-TaskComments=(pmo, 2026-10-04) <day> is booked <N>h against 6h - candidate to move. Change nothing else. Finish with OUTCOME: the overbooked days and the tasks you commented. | pmo | 100 | (claude-code, 2026-10-03 20:58 UTC) (claude-code, 2026-10-03) Completed by the pmo seat itself on Ollama qwen3-coder (no Claude tokens): commented GEN.1, GEN.3, GEN.4, GEN.5 as overbooked-day candidates. Partial: it also found 2026-10-12..14 overbooked but did not comment those. | ||||
| Projects | ctx-5 | CTX.5 - Deferred tool definitions at the MCP door | 2026-10-03T20:15:12.3417481+00:00 | claude-code | CTX.5 | CTX | Deferred tool definitions at the MCP door | 2026-10-20 | 6 | Not Started | Operator 2026-10-03: a total diff view for agents - concise, nothing shown twice, pretrained terms, compressed. Scout: ResearchEnrichment research (context engineering) - JetBrains observation masking (half cost, same quality), Anthropic context editing (-84%) and tool search (-85%, higher accuracy), CSV vs JSON (1/3), TOON, read-dedup pitfalls (stub must say where; hash on content). Baseline from the copilot footer: web question 17,285 tokens, add 4,400. The door lists 125 tools to every seat. Serve core tools plus tool_search; the rest load on demand (the published 85% saving). Screen accuracy with tools/llm-toolcall-probe.py before and after. | claude-code (seat) | ||||||
| Projects | ctx-4 | CTX.4 - Compact results: tables, ids first, full text on demand | 2026-10-03T20:15:12.3009009+00:00 | claude-code | CTX.4 | CTX | Compact results: tables, ids first, full text on demand | 2026-10-12 | 4 | Not Started | Operator 2026-10-03: a total diff view for agents - concise, nothing shown twice, pretrained terms, compressed. Scout: ResearchEnrichment research (context engineering) - JetBrains observation masking (half cost, same quality), Anthropic context editing (-84%) and tool search (-85%, higher accuracy), CSV vs JSON (1/3), TOON, read-dedup pitfalls (stub must say where; hash on content). Baseline from the copilot footer: web question 17,285 tokens, add 4,400. SearchRows, ReadNodes, site/list tools and web search return a header-once table (ids, titles), top-k; detail via ReadNodes / WebFetch; web_fetch default 3,000 chars. Narrative stays key:value. | claude-code (seat) | ||||||
| Projects | ctx-3 | CTX.3 - Change log for agents: GetChanges since a ChangeToken | 2026-10-03T20:15:12.2243825+00:00 | claude-code | CTX.3 | CTX | Change log for agents: GetChanges since a ChangeToken | 2026-10-15 | 6 | Not Started | Operator 2026-10-03: a total diff view for agents - concise, nothing shown twice, pretrained terms, compressed. Scout: ResearchEnrichment research (context engineering) - JetBrains observation masking (half cost, same quality), Anthropic context editing (-84%) and tool search (-85%, higher accuracy), CSV vs JSON (1/3), TOON, read-dedup pitfalls (stub must say where; hash on content). Baseline from the copilot footer: web question 17,285 tokens, add 4,400. SharePoint's change log as the diff view: query_my_tasks / query_my_context accept since=<ChangeToken>; a seat's briefing opens with what changed since its last session, not everything again. Measure tokens per briefing before and after. | claude-code (seat) | ||||||
| Projects | ctx-2 | CTX.2 - Read dedup within an answer, keyed on content hash | 2026-10-03T20:15:12.1668325+00:00 | claude-code | CTX.2 | CTX | Read dedup within an answer, keyed on content hash | 2026-10-08 | 3 | Not Started | Operator 2026-10-03: a total diff view for agents - concise, nothing shown twice, pretrained terms, compressed. Scout: ResearchEnrichment research (context engineering) - JetBrains observation masking (half cost, same quality), Anthropic context editing (-84%) and tool search (-85%, higher accuracy), CSV vs JSON (1/3), TOON, read-dedup pitfalls (stub must say where; hash on content). Baseline from the copilot footer: web question 17,285 tokens, add 4,400. Same node, row or URL inside one answer returns [unchanged since turn N - <where>] when its content hash matches; any change returns the full content. Extends BRAINSTORM.29 (identical read-only calls). | claude-code (seat) | ||||||
| Projects | ctx-1 | CTX.1 - Observation masking in the tool loop | 2026-10-03T20:15:12.1182731+00:00 | claude-code | CTX.1 | CTX | Observation masking in the tool loop | 2026-10-08 | 5 | Not Started | Operator 2026-10-03: a total diff view for agents - concise, nothing shown twice, pretrained terms, compressed. Scout: ResearchEnrichment research (context engineering) - JetBrains observation masking (half cost, same quality), Anthropic context editing (-84%) and tool search (-85%, higher accuracy), CSV vs JSON (1/3), TOON, read-dedup pitfalls (stub must say where; hash on content). Baseline from the copilot footer: web question 17,285 tokens, add 4,400. Tool results older than the last 3 become a one-line stub (tool, turn, title/id, size - shown earlier); a needed fact is fetched again by id. In the chat loop and the phase ToolLoop. Target: the copilot web question under 6,000 tokens with the same answer. | claude-code (seat) | ||||||
| Projects | gov-16 | GOV.16 - Rules once, referenced everywhere: skills and briefings point to the governance pages | 2026-10-03T19:03:22.1580708+00:00 | claude-code | GOV.16 | GOV | Rules once, referenced everywhere | 2026-10-13 | 4 | Not Started | Operator 2026-10-03: generate skills from it so the same info is not everywhere. Deliver: Skill.FollowGovernance (and the seats' and Genesis' briefings) carry a knowledge slot / Query.GovernancePages pointer to the Governance pages (Manifesto, When to use what, Glossary), never the rule text (STAFF.1 progressive disclosure); audit the copies - KB articles, procedure text, part descriptions, the claude-code memory files, skills - and replace each restated rule with a pointer to its page. Measure: one rule edited on the page reaches a seat's next briefing with no other edit. | claude-code (seat) | ||||||
| Projects | tpl-36 | TPL.36 - Import the standard intranet: PnP provisioning templates through Workflow.SiteKit | 2026-10-03T19:00:30.7728982+00:00 | claude-code | TPL.36 | TPL | Import the standard intranet (PnP templates) | 2026-10-16 | 6 | Not Started | Operator 2026-10-03: as much as possible the SharePoint way; intranet things that exist, just import. Scout the PnP originals (SharePoint Look Book, PnP Starter Kit / provisioning templates: news, events, FAQ, people directory, policies and procedures, onboarding, department sites, company home), download them, and import each through Workflow.SiteKit (PnP template to a delta parked for approval). Build only the SPICE difference. Feeds TPL.35 (jumpstart site) and GEN.5 (agency site): their standard parts come from these imports. | claude-code (seat) | ||||||
| Projects | gov-15 | GOV.15 - Site metadata and a duplicate check: purpose, owner, lifecycle on every site | 2026-10-03T18:57:57.4260854+00:00 | claude-code | GOV.15 | GOV | Site metadata and duplicate check | 2026-10-12 | 5 | Not Started | Operator 2026-10-03: we should have metadata of a site's purpose, and a glossary, to prevent wild growth and duplication. Today a site has a free-text Description only (73 of them). Deliver (SharePoint shape: site properties shown on the Site Directory / Sites map): SitePurpose (choice: Hub, Team, Project, Agency, MySite, Demo), SiteOwner (a person or seat), Lifecycle (Active, Archived), backfilled on all sites; AddSiteBlueprint requires purpose and owner and refuses a new site whose purpose sentence and type match an existing site (point to it instead); Genesis checks the catalog and the agencies first (When to use what page, Governance hub). The Sites map filters by purpose and lifecycle. | claude-code (seat) | (claude-code, 2026-10-03 19:03 UTC) (claude-code, 2026-10-03) Operator: templates and instances carry this data, rolled up in governance. Add: a SiteTemplate declares its purpose type, owner role, lifecycle default and version; a site created from it stamps SitePurpose, SiteOwner, Lifecycle, TemplateId, TemplateVersion; the Governance hub gets a Sites roll-up (Content Search web part over the site properties) with flags - no owner, duplicate purpose, archived but active, outdated template version. | |||||
| Projects | gen-5 | GEN.5 - The Agency site template: what an agency is, on its site | 2026-10-03T18:53:59.9023550+00:00 | claude-code | GEN.5 | GEN | Agency site template | 2026-10-09 | 6 | Not Started | Operator 2026-10-03: it should be well defined in the template what an agency does and how it differs from other sites. Today SiteTemplate.MissionDivision holds only Digest and Actions; the definition lives in the Agency part. Deliver the agency site (only what differs from a team site): a Charter page (mission, what it does and does NOT do, one success criterion), Services (its catalog rows, published on the agent card), a work queue (requests dispatched to it, STAFF.2), a Runbook (its workflows and procedures = its M.O.), Crew (members, roles, RACI, fallback seat), a Deliverables library, a Scorecard (the success criterion measured) and Lessons. Default content by a stapled Feature (TPL.34). Genesis (GEN.3) fills it for a new agency; the four core agencies get it on approval. | claude-code (seat) | (pmo, 2026-10-03 20:56 UTC) (pmo, 2026-10-04) 2026-10-09 is booked 12h against 6h - candidate to move. | |||||
| Projects | gen-4 | GEN.4 - Toolmaker: new tools as reviewed scripts that run in the worker's Docker sandbox | 2026-10-03T18:46:46.2089534+00:00 | claude-code | GEN.4 | GEN | Toolmaker: new tools as reviewed scripts that run in the worker's Docker sandbox | 2026-10-14 | 10 | Not Started | Operator asked: is there a toolmaker, can it run on a Docker VM, do we need an open terminal? Answer: no shell on the SPICE host. The spice-worker already runs in a Docker sandbox on .69 with network allowlisted to SPICE and its model. Deliver: a ToolScript part (script + inputs + allowed hosts) authored by Genesis' PartSmith, parked for approval, run by the worker inside its sandbox as a tool (tool_run_script), output back through MCP. Scout first: Agency Swarm ToolCreator, CrewAI custom tools, MCP code sandboxes. | claude-code (seat) | (pmo, 2026-10-03 20:56 UTC) (pmo, 2026-10-04) 2026-10-14 is booked 18h against 6h - candidate to move. | |||||
| Projects | gen-3 | GEN.3 - Genesis builds real crews and tools, not empty divisions | 2026-10-03T18:46:46.1501016+00:00 | claude-code | GEN.3 | GEN | Genesis builds real crews and tools, not empty divisions | 2026-10-08 | 8 | Not Started | Today Phase.GenesisCreateCrew may only attach crew Features that already exist, so LessonDesk and CatalogConformanceDesk got a site and the generic Feature.AgencyBasics, while GenesisCreateParts' skills and tools are listed and never created. Deliver: the delta also carries AddActorProfile, AddSkill, AddPhase, AddWorkflow, AddFeature for the new crew (the doors exist), validated, parked as ONE approval; a tool that is new code is a GEN.4 request, not a fake part. Check the two existing desks: fill or retire them. | claude-code (seat) | (pmo, 2026-10-03 20:56 UTC) (pmo, 2026-10-04) 2026-10-08 is booked 15h against 6h - candidate to move. | |||||
| Projects | gen-2 | GEN.2 - Genesis asks before it designs: a questioning phase and a scope verdict | 2026-10-03T18:46:46.0986883+00:00 | claude-code | GEN.2 | GEN | Genesis asks before it designs: a questioning phase and a scope verdict | 2026-10-07 | 5 | Not Started | Operator 2026-10-03 (made these, do they function?): Agency Swarm's GenesisCEO questions the requester until the mission is clear; SPICE's Phase.GenesisDesign designs straight from the request text. Deliver: a first Task step (TPL.17 durable task) that files the open questions to the requester as a Workflow Task and parks until answered; then a scope verdict (snippet / small project / full division, the intake modes) - only a full division continues to design. Measure on a real request. | claude-code (seat) | ||||||
| Projects | gen-1 | GEN.1 - Agency template gallery: Genesis starts from a proven agency (Agency Swarm, CrewAI) instead of from scratch | 2026-10-03T18:42:24.2754562+00:00 | claude-code | GEN.1 | GEN | Agency template gallery for Genesis | 2026-10-06 | 6 | Not Started | Operator 2026-10-03: an agency site that makes other agencies, like Agency Swarm or CrewAI. It exists: Agency.Genesis on /sites/Genesis (design, approval, MissionDivision site + crew; built LessonDesk and CatalogConformanceDesk). Missing: templates to start from. Import as agency templates on a gallery (SharePoint Solution Gallery shape), each: purpose, crew by role (ActorProfile Goal/Backstory), comm graph, the workflow, the tools it needs: C:/Users/erike/ai/agency-swarm-lab (CodeGuardiansAgency, CodeSolutionAgency, MetaMarkAgency, OpenSourceSwarm, WebDevCrafters) and C:/Users/erike/ai/crewAI-examples (game-builder-crew, instagram_post, job-posting, landing_page_generator, markdown_validator, prep-for-a-meeting, screenplay_writer, stock_analysis, trip_planner, starter_template). Phase.GenesisDesign reads the gallery first and names the template it adapted. Import through the provisioning doors (AddAgency, AddActorProfile, AddWorkflow), parked for approval - not by hand. Then: a runtime for the built crews (STAFF.8, GOV.2) and direct routing of new-division requests (INTAKE.13). | claude-code (seat) | (pmo, 2026-10-03 20:56 UTC) (pmo, 2026-10-04) 2026-10-06 is booked 8h against 6h - candidate to move. | |||||
| Projects | infra-8 | INFRA.8 - Term Store Management tool and document check-in/check-out | 2026-10-03T18:03:18.1237615+00:00 | claude-code | INFRA.8 | INFRA | Term Store Management tool and document check-in/check-out | 2026-11-12 | 8 | Not Started | Fit-gap 2026-10-03 (Wisdom: SPICE versus enterprise SharePoint infrastructure), rank 8 of 8 by value for a human operator. PMO: plan it (owner, date) in the next planning pass; build through the provisioning doors where data, by hand only for engine seams. | claude-code (seat) | (pmo, 2026-10-03 18:06 UTC) (pmo, 2026-10-03) Last in order; 8h on 11-10..11-12, inside the 11-15 limit. | |||||
| Projects | infra-7 | INFRA.7 - Quotas and usage analytics per site collection | 2026-10-03T18:03:18.1091839+00:00 | claude-code | INFRA.7 | INFRA | Quotas and usage analytics per site collection | 2026-11-10 | 6 | Not Started | Fit-gap 2026-10-03 (Wisdom: SPICE versus enterprise SharePoint infrastructure), rank 7 of 8 by value for a human operator. PMO: plan it (owner, date) in the next planning pass; build through the provisioning doors where data, by hand only for engine seams. | claude-code (seat) | (pmo, 2026-10-03 18:06 UTC) (pmo, 2026-10-03) Seventh; 6h after the 11-05..11-09 booked rows. | |||||
| Projects | infra-6 | INFRA.6 - Access requests and sharing: request access, approve, the requester added to the right group | 2026-10-03T18:03:18.0934692+00:00 | claude-code | INFRA.6 | INFRA | Access requests and sharing: request access, approve, the requester added to the right group | 2026-11-06 | 6 | Not Started | Fit-gap 2026-10-03 (Wisdom: SPICE versus enterprise SharePoint infrastructure), rank 6 of 8 by value for a human operator. PMO: plan it (owner, date) in the next planning pass; build through the provisioning doors where data, by hand only for engine seams. | claude-code (seat) | (pmo, 2026-10-03 18:06 UTC) (pmo, 2026-10-03) Sixth; 6h on the rest of 11-03 and 11-06. | |||||
| Projects | infra-5 | INFRA.5 - Audit log report page, then holds and eDiscovery | 2026-10-03T18:03:18.0087640+00:00 | claude-code | INFRA.5 | INFRA | Audit log report page, then holds and eDiscovery | 2026-11-03 | 8 | Not Started | Fit-gap 2026-10-03 (Wisdom: SPICE versus enterprise SharePoint infrastructure), rank 5 of 8 by value for a human operator. PMO: plan it (owner, date) in the next planning pass; build through the provisioning doors where data, by hand only for engine seams. | claude-code (seat) | (pmo, 2026-10-03 18:06 UTC) (pmo, 2026-10-03) Fifth; 8h on free days 10-28 and 11-03. | |||||
| Projects | infra-4 | INFRA.4 - Service applications: start, stop and configure from Central Administration | 2026-10-03T18:03:17.9906785+00:00 | claude-code | INFRA.4 | INFRA | Service applications: start, stop and configure from Central Administration | 2026-10-23 | 5 | Not Started | Fit-gap 2026-10-03 (Wisdom: SPICE versus enterprise SharePoint infrastructure), rank 4 of 8 by value for a human operator. PMO: plan it (owner, date) in the next planning pass; build through the provisioning doors where data, by hand only for engine seams. | claude-code (seat) | (pmo, 2026-10-03 18:06 UTC) (pmo, 2026-10-03) Fourth; 5h on free hours of 10-23, after INFRA.3. | |||||
| Projects | infra-3 | INFRA.3 - Health Analyzer rules: disk, database, expired secrets, failing services - with Repair | 2026-10-03T18:03:17.9731984+00:00 | claude-code | INFRA.3 | INFRA | Health Analyzer rules: disk, database, expired secrets, failing services - with Repair | 2026-10-23 | 6 | Not Started | Fit-gap 2026-10-03 (Wisdom: SPICE versus enterprise SharePoint infrastructure), rank 3 of 8 by value for a human operator. PMO: plan it (owner, date) in the next planning pass; build through the provisioning doors where data, by hand only for engine seams. | claude-code (seat) | (pmo, 2026-10-03 18:06 UTC) (pmo, 2026-10-03) Third; 6h on the sparse days 10-13..10-23. | |||||
| Projects | infra-2 | INFRA.2 - Timer job control: Run now, enable or disable, last result and error per job | 2026-10-03T18:03:17.9457388+00:00 | claude-code | INFRA.2 | INFRA | Timer job control: Run now, enable or disable, last result and error per job | 2026-10-03 | 4 | Completed | Fit-gap 2026-10-03 (Wisdom: SPICE versus enterprise SharePoint infrastructure), rank 2 of 8 by value for a human operator. PMO: plan it (owner, date) in the next planning pass; build through the provisioning doors where data, by hand only for engine seams. | claude-code (seat) | 100 | (pmo, 2026-10-03 18:06 UTC) (pmo, 2026-10-03) Built now per claude-code; 4h fits today. (claude-code, 2026-10-03 18:13 UTC) (claude-code, 2026-10-03) Shipped 11566ce, 4351/4351, verified live: Run now on Schedule.ContentOrganizerProcessing -> succeeded with its message on /board/schedule; Enable/Disable on the job's state row override Scope.Scheduler without a restart; anonymous LAN caller refused (403). Not done: a job history list (only the last run is kept; workflow runs are in Workflow History). | ||||
| Projects | infra-1 | INFRA.1 - Backup and restore in Central Administration: site and list backup, restore, export | 2026-10-03T18:03:17.9025434+00:00 | claude-code | INFRA.1 | INFRA | Backup and restore in Central Administration: site and list backup, restore, export | 2026-10-13 | 8 | Not Started | Fit-gap 2026-10-03 (Wisdom: SPICE versus enterprise SharePoint infrastructure), rank 1 of 8 by value for a human operator. PMO: plan it (owner, date) in the next planning pass; build through the provisioning doors where data, by hand only for engine seams. | claude-code (seat) | (pmo, 2026-10-03 18:06 UTC) (pmo, 2026-10-03) Next in INFRA order; 8h on free hours 10-03..10-13 around booked rows. | |||||
| Projects | pmo-2 | PMO.2 - What the PMO seat missed at the MCP door | 2026-10-03T17:41:21.9880523+00:00 | claude-code | PMO.2 | PMO | PMO door friction from the first planning pass | 2026-10-12 | 3 | Completed | Found by the pmo seat on PMO.1 (2026-10-03): (1) no stored query for open tasks of a project with an empty owner or date - it pulled every row and filtered by hand: add Query.UnplannedTasks (a SavedQuery, served as query_unplanned_tasks); (2) spice.py rows truncates cells, so it could not read whole comments to judge dependencies; (3) a seat's My Site Tasks has no TaskComments column, so its summary went into TaskOutcome - the delegated task should carry the comment field (CrewTask). | claude-code (seat) | 100 | (claude-code, 2026-10-03 20:58 UTC) (claude-code, 2026-10-03) Done by running the pmo seat on qwen3-coder (Ollama .69) until it worked: Query.AssignedWork (owner + due window; 5,985 chars where a dump was 54,675), MCP results compact JSON, tool_update_list_item takes itemId/key/id and fields as field-X, a fields object or bare column names (REST shape). Commits 3574d38 and after. | ||||
| Projects | tpl-35 | TPL.35 - Jumpstart site: one template with everything we already thought of, ready to use | 2026-10-03T17:40:56.8491759+00:00 | claude-code | TPL.35 | TPL | Jumpstart site template | 2026-10-23 | 8 | Waiting on someone else | Operator 2026-10-03: a template site where all we thought of is in and ready - a jumpstart. SharePoint shape: a site template plus stapled features with default content (TPL.34 ListInstance Data). Contents to gather: the Project Site lists and default rows (stage tasks, anti-pattern risks, links), the enterprise groups nested (Board Owners, the team Members, Agents/People Visitors), a Gantt and Calendar view, the service contract (Reception catalog row), the onboarding procedure, a Meeting Workspace (STAFF.9) when it exists, Mermaid flows on the home page. PMO: list the candidates as a catalog the operator cherry-picks (Fit/Gap), then the builders assemble - through the doors of HARNESS.20, not by hand. Outcome (pmo, 2026-10-03): [attempt 1/3] worker error: HTTP Error 500: Internal Server Error Outcome (pmo, 2026-10-03): unknown: Unable to complete the task as the tool requires at least one field to be updated, and the 'TaskOutcome' field was not recognized as a valid update field in the current context. | pmo | ||||||
| Projects | harness-20 | HARNESS.20 - No hand edits of the spine: the missing provisioning doors | 2026-10-03T17:40:56.3894696+00:00 | claude-code | HARNESS.20 | HARNESS | No hand edits of the spine: the missing provisioning doors | 2026-10-09 | 6 | Not Started | Operator 2026-10-03: why do you still edit things manually? Audit of this session: Risk CT and fields (AddContentType and AddFieldDefinition existed - should have been a parked delta), Feature.ProjectSiteDefaults (AddFeature existed), Workflow.ReceiveAndDispatch steps, SiteTemplate.ProjectSite lists and the Directory enterprise groups had NO door. Deliver the doors: UpdatePart (replace an existing part by Id, validated against the XSD, the old version kept), AddSiteTemplate, AddSiteGroup (create a group on a site; AddGroupMember fills it). Rule after this: parts and site data go through a ProvisioningDelta parked for approval; hand edits only for C#, XSD and XSLT engine work. | claude-code (seat) | (claude-code, 2026-10-03 19:40 UTC) (claude-code, 2026-10-03) Hand edit, no door: Actor.MapCopilot Goal - adding comes first (UpdatePart needed). | |||||
| Projects | pmo-1 | PMO.1 - Planning pass: owners, dates and estimates for STAFF, INTAKE, GOV and BRAINSTORM | 2026-10-03T17:33:55.4690087+00:00 | claude-code | PMO.1 | PMO | Planning pass for the focus projects | 2026-10-04 | 2 | Completed | Give every open task in STAFF, INTAKE, GOV and BRAINSTORM an owner, a DueDate and an estimate, by the planning rules in your SHAREPOINT.md (my-pmo Agent Assets). Merge with tool_update_list_item; never change another owner's Status or Body. Finish with one summary comment here: planned, not planned (why), decisions needed from the operator. Check the result on /sites/Projects/Lists/ProjectTasks?view=gantt. Outcome (pmo, 2026-10-03): Planned 39 open rows in STAFF/INTAKE/GOV/BRAINSTORM, 2026-10-05..2026-11-09; blockers and operator decisions in Projects/ProjectTasks pmo-1 TaskComments. | pmo | 100 | (pmo, 2026-10-03 17:36 UTC) (pmo, 2026-10-03) Planned 41 open rows: STAFF 8 (due 10-06..11-04), INTAKE 13 (10-05..10-30; intake-13 was pre-planned), GOV 12 (10-05..11-05; gov-12b date is a target), BRAINSTORM 6 (10-06..11-10). Range 2026-10-05..2026-11-10; claude-code at 6h/day. Not firmly plannable: gov-12b (operator approval parked), gov-12d/12e (Waiting on someone else; sequenced after 12c/12d, 12e also needs a runtime), brainstorm-6/gov-2 (operator). Decisions: approve gov-12b roster; run gov-2 seat; STUDIOS session for brainstorm-6; intake-3 (12h) and intake-10 (10h) are large - split?; claude-code is loaded ~5 weeks, consider other owners. | ||||
| Projects | intake-14 | INTAKE.14 - The agent card is the service contract | 2026-10-03T17:27:32.4960092+00:00 | claude-code | INTAKE.14 | INTAKE | The agent card is the service contract | 2026-10-03 | 2 | Completed | Operator 2026-10-03: a contract service so agents work together fluently. Delivered 756e48d: /.well-known/agent.json publishes the Reception service catalog as services, each with the MCP call that requests it (tool_publish_to_list, field-ServiceRef), plus access (seats, people, secrets, health, profile). Catalog rows added: onboard-agent-seat, site-access, seat-secret. Next: INTAKE.13 routes a named service deterministically. | claude-code (seat) | 100 | |||||
| Projects | staff-9 | STAFF.9 - Meeting Workspace: a meeting room where the operator meets the crew with all context | 2026-10-03T15:10:19.9106154+00:00 | claude-code | STAFF.9 | STAFF | Meeting Workspace: meet the crew with all context | 2026-11-04 | 8 | Not Started | Operator 2026-10-03: a chat like this one should happen in a meeting/project room with all context and the people needed. SharePoint: the Meeting Workspace (MPS#0) - Agenda, Attendees, Objectives, Decisions, Document Library, linked from a Calendar event. SPICE delta: Attendees may be agents (CEO, PMO lead, crew) as well as people; the room's chat runs each agent with the linked project site as context; decisions and action items land as rows (Decisions list, Tasks on the project site); reuse Workflow.PlanningMeeting / SteeringMeeting. Depends on STAFF.7 (chat with a chosen actor) and GOV.13 (sign-in). Plan before build: brainstorm on the map first. | claude-code (seat) | (pmo, 2026-10-03 17:36 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-11-04 at 6h/day: meeting workspace, after staff-7 chat | |||||
| Projects | staff-8 | STAFF.8 - Research and recommend: provisioning crews and agents in SPICE | 2026-10-03T15:10:19.5746965+00:00 | claude-code | STAFF.8 | STAFF | Research and recommend: provisioning crews and agents | 2026-11-02 | 6 | Not Started | Operator 2026-10-03: research how crews and agents are best provisioned, then recommend one implementation for SPICE. Scout: Paperclip hire_agent and org chart, CrewAI crews vs flows, MetaGPT SOP roles, Project Server Enterprise Resource Pool (generic to named resources), SharePoint provisioning (site request, feature activation). Map onto what exists: Genesis/AutoGenesis (AddAgency parked for approval), Agency/ActorProfile parts, Directory/People seats, SeatPool, Feature BindSkill, Scope caps. Deliverable: a Pending Wisdom page with the recommendation (only what differs from SharePoint) plus backlog rows; feeds STAFF.3 (hire when needed) and the PMO crew model (KB 'How SPICE staffs a project'). | claude-code (seat) | (pmo, 2026-10-03 17:36 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-11-02 at 6h/day: research, after the build items | |||||
| Projects | gov-14 | GOV.14 - The My Site is everyone's default landing page, with the basic links, provisioned by feature activation | 2026-10-03T14:04:51.6655770+00:00 | claude-code | GOV.14 | GOV | My Site as default landing page with basic links (feature activation) | 4 | Not Started | Operator 2026-10-03. SharePoint shape: the My Site host is a user's landing page; a Feature (Scope=Web) stapled to SiteTemplate.MySite (FeatureSiteTemplateAssociation) provisions a Links list (ListTemplate Links, 103) with default rows and a links web part on the home page - activated on every existing My Site and on each new one. Basic links: Brainstorm, Projects (my tasks), Reception (make a request), Governance hub (approvals), Wisdom, Search Center, the Board. After GOV.13 sign-in, / redirects a signed-in person to /sites/my-{user}; seats unchanged. | ||||||||
| Projects | gov-13 | GOV.13 - Browser sign-in through Authentik (OIDC), the SharePoint trusted identity provider way | 2026-10-03T13:57:36.3987183+00:00 | claude-code | GOV.13 | GOV | Browser sign-in through Authentik (OIDC) | 2026-10-06 | 4 | In Progress | Operator 2026-10-03 asked Authentik or our own. Recommended: Authentik - SharePoint on-prem signed users in through a trusted identity token issuer (ADFS/SAML); Authentik is that issuer. ASP.NET Core has OpenID Connect built in (cookie + OIDC handler), so no own password store. Map the OIDC user (email/username) to the Directory/People row; loopback keeps System Account, Bearer seats unchanged. Phone on the LAN then signs in at http://192.168.123.153:5198. Needs: an Authentik application/provider (client id, secret, redirect /signin-oidc), secret in user-secrets. Own Forms-Based Auth only if Authentik is not reachable from the LAN. | claude-code (seat) | 80 | (claude-code, 2026-10-03 17:04 UTC) (claude-code, 2026-10-03) Built 22cedba: /_layouts/15/Authenticate.aspx and SignOut.aspx on the existing SPICE.Plugins.Auth.Oidc; a signed-in person resolves as the operator under their own name; forwarded headers trusted from 192.168.123.69. Verified: sign-in 302 to Authentik with PKCE. Waiting: the client secret, sealed by system-steward to the SPICE host key, into user-secrets Auth:ClientSecret; then spice.angelsworks.org behind the proxy for phones. (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) owner was empty; in progress, due 2026-10-06: in progress, built 22cedba; finish and verify sign-in | ||||
| Projects | staff-7 | STAFF.7 - Chat with the CEO (and any crew actor) beside the brainstorm | 2026-10-03T13:57:36.0739206+00:00 | claude-code | STAFF.7 | STAFF | Chat with the CEO (and any crew actor) beside the brainstorm | 2026-10-21 | 4 | Not Started | Operator 2026-10-03: a CEO chat somewhere, to talk with the company instead of with Claude Code. The Brainstorm map chat is fixed to Actor.MapCopilot. Deliver: an actor picker on the chat (or a chat web part on the Board/Governance page) that runs the chosen actor (CEO = Actor.ChiefStrategyOfficer, GOV.12) on the local model with the map selection as context; the CEO needs skills first (GOV.12d: empty SkillCatalog). Conversation stays a row on the site's Conversations list. | claude-code (seat) | (pmo, 2026-10-03 17:36 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-21 at 6h/day: chat with CEO beside brainstorm | |||||
| Projects | staff-6 | STAFF.6 - Agents work a task only through its form: the payload rides the task, each party edits only its own part | 2026-10-02T16:39:25.1218677+00:00 | claude-code | STAFF.6 | STAFF | Agents work a task only through its form: the payload rides the task, each party edits only its own part | 2026-10-21 | 6 | Not Started | Operator 2026-10-02: use the payload task with InfoPath so agents only operate through forms and can only change their part, or adjust rights on fields. SharePoint shape: a 2010 workflow's task edit form is an InfoPath form bound to the task content type; the request travels in the task (ExtendedProperties), the assignee fills only the answer fields. SPICE today: Workflow Task rows carry WorkflowData JSON; FormTemplate (TPL.24) rules are enforced at every write door, but FieldRef ReadOnly is only a render hint in FormView.xslt - no door refuses a write to it. Deliver: (1) Form.FulfilRequest bound to the dispatch task - the request's Subject/Body/DecisionSupported copied in read-only, the agency's Deliverable, Outcome, Comments editable; (2) ReadOnly enforced server-side in ItemUpdateService/AddListItem/REST for the task content type (the workflow engine writes as Manager, the assignee cannot); (3) optionally per-role editability (InfoPath views per role) - ask first, SharePoint has no field-level permissions natively. Found by the full-circle test: req-news-to-tiktok reached Agency.Content as task wftask-run-9c05840afce1-1-agency-content. | claude-code (seat) | (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-21 at 6h/day: task form payload, after staff-5 composer | |||||
| Projects | harness-19 | HARNESS.19 - spice rows ignores --limit with --order; duplicate research rows flood reads | 2026-10-02T16:14:10.1198161+00:00 | claude-code | HARNESS.19 | HARNESS | spice rows ignores --limit with --order; duplicate research rows flood reads | 2 | Not Started | Found 2026-10-02: python tools/spice.py rows ResearchEnrichment/ResearchReports --order Modified- --limit 3 returned about 150 rows; the list holds dozens of near-identical AI agent frameworks / Captains intel rows (a repeated schedule writing new keys). Fix the client limit, and have the schedule upsert by topic and date instead of appending - noise an agent reads is context it pays for (INTAKE.5). | ||||||||
| Projects | intake-13 | INTAKE.13 - The dispatcher honours a stated ServiceRef | 2026-10-02T16:14:09.9149608+00:00 | claude-code | INTAKE.13 | INTAKE | The dispatcher honours a stated ServiceRef | 2026-10-05 | 3 | Not Started | Found 2026-10-02: req-research-agent-staffing was filed WITH ServiceRef=deep-research, yet Phase.DispatchRequest (an LLM phase) routed it to Agency.Content, and nothing produced a report. A request that names a catalog service goes to that service owner deterministically (a DecisionTable rule before the model is asked); the model only routes requests that name none. Pair with INTAKE.6 (a dispatched request reaches a fulfiller). | claude-code (seat) | (claude-code, 2026-10-03 17:15 UTC) (claude-code, 2026-10-03) Now on the fluent path: the agent card publishes the catalog and peers file requests with ServiceRef (756e48d). Design for the deterministic route, SharePoint Designer's Workflow Lookup: a SetField step whose value is looked up in another list (Data source Reception/ServiceCatalog, field ServiceOwner, find the item where its key = Current Item:ServiceRef), applied only when it finds a row, so it overrides the model's guess and never blanks a target. | |||||
| Projects | staff-5 | STAFF.5 - One composer: phase briefings read the same layers | 2026-10-02T15:56:47.2243855+00:00 | claude-code | STAFF.5 | STAFF | One composer: phase briefings read the same layers | 2026-10-20 | 4 | Not Started | BriefingBuilder (phase briefings, hard-coded in C#) reads the Scope.SeatBriefing layer settings, so Wisdom, clipping and budgets are chosen once for seats and phases alike. Later: Field.MonthlyBudget on Person with alert and pause (ask first). Risk: phase briefing regressions across the XSLT tests. Size L. | claude-code (seat) | (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-20 at 6h/day: after staff-4: one composer reads same layers (est by PMO) | |||||
| Projects | staff-4 | STAFF.4 - JIT briefing for every hire: mission topic, what differs, measured | 2026-10-02T15:56:47.0063499+00:00 | claude-code | STAFF.4 | STAFF | JIT briefing for every hire: mission topic, what differs, measured | 2026-10-19 | 4 | Not Started | Tool.ComposeBriefing is the JIT algorithm (Scope.SeatBriefing layers with char budgets, dedupe, quota scaling, topic on demand) but only Claude Code uses it: the worker harness starts from it too (GOV.9); add a mission-topic layer and a 'what differs from SharePoint' Wisdom layer; measure the briefing per actor and set a ceiling (INTAKE.5). Size M. | claude-code (seat) | (claude-code, 2026-10-02 16:13 UTC) Research staff-what-works-2026-10-02: Agent Skills three levels (name+description, body when relevant, files on demand) [17]; context rot - smallest high-signal set, references loaded at runtime [18]; deferred tool loading cut about 77K to 8.7K tokens [19]. A briefing is an index plus pointers. (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-19 at 6h/day: JIT briefing, research in comments | |||||
| Projects | staff-3 | STAFF.3 - Hire when needed, retire when idle (Paperclip's hire_agent, Project Server's generic resources) | 2026-10-02T15:56:46.7869354+00:00 | claude-code | STAFF.3 | STAFF | Hire when needed, retire when idle (Paperclip's hire_agent, Project Server's generic resources) | 2026-10-16 | 6 | Not Started | Generalise AutoGenesis' rank-and-park leg (it already parks AddAgency for unmet demand, capped by AutoGenesis.IdleBornCap) to hires: demand with no fulfiller (unrouted requests, repeat subjects INTAKE.2, work a free agent cannot place INTAKE.4) parks an actor hire from a role template (Scope.GeneratedOperator) - a seat hire only when the template names a running harness. Caps and IdleDays in Scope.Staffing (a scope, not a new kind); an idle hire is retired the same gated way. Approval by policy (INTAKE.12, Jev). Ask first: the approval type column (GOV.12f). Size M. | claude-code (seat) | (claude-code, 2026-10-02 16:13 UTC) Research staff-what-works-2026-10-02: a hire = an ApprovalRequest whose approval provisions it (Paperclip hire_agent [1][2]); the request names the capability gap and is checked against existing roles; caps as thresholds - hires per site, delegation depth, monthly spend (sprawl cost 47K USD in 11 days elsewhere [4]); role before person (Project Server generic resource = a Directory SiteGroup [10]); retire on no assignments AND no runs for N days, archive never delete (KEDA scale-to-zero pitfall [8]). (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-16 at 6h/day: hire when needed, research already in comments | |||||
| Projects | staff-2 | STAFF.2 - A dispatched request reaches its fulfiller, and the default roster | 2026-10-02T15:56:46.5687271+00:00 | claude-code | STAFF.2 | STAFF | A dispatched request reaches its fulfiller, and the default roster | Completed | INTAKE.6 by reuse: Phase.DispatchRequest's delta also files a Task on the target agency's site (the SeatFallback assign shape); with no fulfiller it parks a hire instead of NeedsAgency (STAFF.3). The default staff (GOV.12: Board = operator, C-suite and operating roles on existing actors, reports-to) declared as one gated delta of People rows (GOV.12b). Served test: drop a request, see the Task land. Size S-M. | 100 | (claude-code, 2026-10-02 16:28 UTC) (claude-code, 2026-10-02) Plan: reuse TPL.17 - Workflow.ReceiveAndDispatch gets Step Kind=Task Assignees={IncomingRequestTarget} then SetField IncomingRequestStatus=Resolved, FallbackSeat=claude-code; Reception gets a Tasks list. Pivot from the SeatFallback shape: the task lands on the desk's own Tasks list assigned to the agency (SharePoint fixes the task list at association), not on the agency's site. Engine fixes: a task that does not land makes Park fail; an unassignable step is not stamped Completed. Honest limit: the task is visible, not yet acted on - INTAKE.6 gives agencies a receiver. Roster delta stays on GOV.12b. (claude-code, 2026-10-02 16:39 UTC) (claude-code, 2026-10-02) Shipped 011ec28, 4345/4345. Live: req-news-to-tiktok dispatched by the local model to Agency.Content; task wftask-run-9c05840afce1-1-agency-content landed on Reception/Tasks; request In Progress. Limit: the task is visible, nothing acts on it yet (INTAKE.6, STAFF.6). Roster delta moved to GOV.12b. | |||||||
| Projects | staff-1 | STAFF.1 - Fix and delete: the KB detail link, the Requests slot, dead context code | 2026-10-02T15:56:46.3554284+00:00 | claude-code | STAFF.1 | STAFF | Fix and delete: the KB detail link, the Requests slot, dead context code | 2026-10-06 | 2 | Not Started | Architect 2026-10-02: BriefingBuilder.cs:290 emits KnowledgeById(...) slots no resolver serves - add Query.KnowledgeById (clone Query.KnowledgeByTag with an Id filter) and point the slot at it, with a test that FETCHES the slot; Phase.DispatchRequest's ListView RowLimit=10 should be limit=10 (ListViewSlotResolver reads limit); delete JITContextInjector.cs (unreferenced), ContextManager's key-value half (only the dead injector used it), the duplicate AddSingleton<ContextManager> (SpiceInfrastructureExtensions.cs:47) and BriefingBuilder's never-set User branch. Size S. | claude-code (seat) | (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-06 at 6h/day: quick fix/delete cleanup, unblocks staff chain | |||||
| Projects | intake-12 | INTAKE.12 - Approval by operating mode: Startup and Live, an approval policy table, a steward seat | 2026-10-02T15:55:36.0536995+00:00 | claude-code | INTAKE.12 | INTAKE | Approval by operating mode: Startup and Live, an approval policy table, a steward seat | 2026-10-14 | 8 | Not Started | Operator 2026-10-02: "can we have autoapprove on modes, phases like startup and not live mode, or have a seat with a dev agent?". Design (data first, no new part kind): (1) the farm operating mode is a Setting (Startup | Live) the Board switches; (2) the approval policy is a DecisionTable (DMN, like DT.AgencyTriage): ApprovalType (page, build, skill, hire, strategy, budget, research spend - GOV.12f) x Mode -> AutoApprove | Steward | Board; (3) an Approvals steward seat whose agent files the ApprovalVerdict the policy hands it - VerdictQuorumListener already counts an agent verdict; (4) one engine seam: when an ApprovalRequest is parked, evaluate the table - AutoApprove writes a System verdict marked auto-approved in Startup (audited), Steward files a task on the steward seat, Board waits for the operator. Page-only changes auto-approve in both modes, so scheduled runs (RT-VIZ) never stall (INTAKE.7). Depends on GOV.12f (the ApprovalType column). Fits the staffing + JIT context module (design pending). | claude-code (seat) | (claude-code, 2026-10-02 15:56 UTC) Operator 2026-10-02: the steward is the decision model Jev (Provider.Jev, TypeSafe Jev 1.13: noul / choice / score, never text). Reuse SPWAY.8c: a DecisionTable Input can be decide(question, options) evaluated by Jev, as DT.AgencyTriage decides a request Kind. DT.ApprovalPolicy: Inputs ApprovalType, Mode (Startup|Live) and Risk = decide(How risky is this change for the live sites?, low|medium|high); rules e.g. page -> AutoApprove; Startup and risk low|medium -> AutoApprove; Live and build and risk low -> AutoApprove; hire, strategy, budget or risk high -> Board. Every auto-approval is a System verdict naming the rule and Jev probability (audited). Cost about 0.04 USD per million tokens. (claude-code, 2026-10-02 15:59 UTC) Operator 2026-10-02 correction: the Approvals steward is a LIVE staff member on the infra production server (the AngelsWorks hub side), NOT a seat or actor inside SPICE. So DT.ApprovalPolicy's Steward outcome hands the ApprovalRequest to the hub through a declared Connector (the Connector.HqHeartbeat precedent: no C# per endpoint) as an inbox handoff carrying the request link and the proposed delta summary; the steward answers through SPICE's MCP door with its own app principal, filing the ApprovalVerdict that VerdictQuorumListener counts. Jev (Provider.Jev) stays the in-SPICE decision for AutoApprove; Board stays the operator. (claude-code, 2026-10-02 16:13 UTC) Research staff-what-works-2026-10-02: ITIL standard changes are pre-authorised [12]; DORA - external approval does not lower change failure rate, teams relying on it were 2.6x more likely low performers [13]: default auto-approve plus an audit row, humans only for irreversible or high-blast-radius changes. Jev grades, the table decides. The steward hand-off = an escalation chain with a due date and default-deny expiry, verdict by callback [11][15][16]. (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-14 at 6h/day: approval policy by operating mode; operator decision 2026-10-02 already recorded; intake-7 depends on it | |||||
| Projects | intake-11 | INTAKE.11 - A refused write says why: tool_publish_to_list returned an empty error | 2026-10-02T15:46:26.1801168+00:00 | claude-code | INTAKE.11 | INTAKE | A refused write says why: tool_publish_to_list returned an empty error | 2026-10-07 | 2 | Not Started | Found 2026-10-02: publishing a ProjectTask with a TaskTitle over the column's length failed with 'Publishing to Projects/ProjectTasks failed:' and no reason - an agent cannot correct what it is not told. Every refusal names the field and the limit (the per-field errors the REST create returns). | claude-code (seat) | (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-07 at 6h/day: small bug fix, refused write must say why | |||||
| Projects | intake-10 | INTAKE.10 - Genesis builds with builder ops and tries on Demo | 2026-10-02T15:46:25.9435459+00:00 | claude-code | INTAKE.10 | INTAKE | Genesis builds with builder ops and tries on Demo | 2026-10-30 | 10 | Not Started | Workflow.GenesisBuildDivision's CreateParts/CreateCrew phases emit builder ops (AddFieldDefinition, AddContentType, AddWorkflow, AddEventReceiver, AddFormTemplate, AddView) in its gated delta instead of free-form part fragments, reuse site/list/workflow templates, and dry-run on the Demo site collection. Touches the Reception->Genesis hand-off (ER.OnRequestDispatchedToGenesis). Size M-L. | claude-code (seat) | (pmo, 2026-10-03 17:36 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-30 at 6h/day: Genesis builder ops, after intake-3 and intake-7 | |||||
| Projects | intake-9 | INTAKE.9 - Retire the legacy factory crew: Parts Factory, RT-MAKE-PAGE, the smith actors | 2026-10-02T15:46:25.5978593+00:00 | claude-code | INTAKE.9 | INTAKE | Retire the legacy factory crew: Parts Factory, RT-MAKE-PAGE, the smith actors | 2026-10-07 | 2 | Not Started | After INTAKE.7: delete Agency.PartsFactory, RT-MAKE-PAGE (SeedDeltas/PartsFactoryLine.xml), Actor.PageSmith, Actor.PipelineWright, Actor.OperatorSmith - crews with no tools behind them; Actor.SiteBuilder (Skill.SiteProvisioning + Skill.BuildSharePointFeatures) is the builder. Keep Scope.PageTemplates (data, used by publishing). Size S. | claude-code (seat) | (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-07 at 6h/day: small retire/cleanup | |||||
| Projects | intake-8 | INTAKE.8 - RequestCapability: template first, gaps as New Features | 2026-10-02T15:46:25.3848080+00:00 | claude-code | INTAKE.8 | INTAKE | RequestCapability: template first, gaps as New Features | 2026-10-12 | 3 | Not Started | Before composing a pipeline, check the site/list template catalogue (Tool.ListSiteTemplates catalogue=true) - a feature need is usually a template or a feature, not a pipeline; repoint the 'request a new block' branch from /gaps and /forge to Tool.FileIssue (New Feature on Projects/Issues). Size S. | claude-code (seat) | (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-12 at 6h/day: RequestCapability template, small | |||||
| Projects | intake-7 | INTAKE.7 - Builders park, never write: MakePage, BuildPipeline, RequestCapability | 2026-10-02T15:46:25.1725866+00:00 | claude-code | INTAKE.7 | INTAKE | Builders park, never write: MakePage, BuildPipeline, RequestCapability | 2026-10-15 | 6 | Not Started | Factory analysis 2026-10-02: Tool.MakePage (MakePageToolInvoker.cs:106), Tool.BuildPipeline (BuildPipelineToolInvoker.cs:183) and RequestCapability's build branch call applier.Apply(dryRun:false) as Manager - an immediate write with no approval, unlike Tool.RequestSite / BuildDivision / ProposeSiteKit which park an ApprovalRequest (BuildParkDelta). Make them park too. Decide first: cron standing orders (RT-VIZ calls MakePage) would stall behind approval - auto-approve the page kind, or keep a direct path for scheduled runs. Size M. | claude-code (seat) | (claude-code, 2026-10-02 15:55 UTC) Operator 2026-10-02: auto-approve, by operating mode - the approval policy of INTAKE.12 decides (page-only changes auto-approve in every mode, so scheduled runs never stall). (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-15 at 6h/day: after intake-12 (approval policy decides auto-approve) | |||||
| Projects | intake-5 | INTAKE.5 - Progressive disclosure and teaching cost in briefings and skills | 2026-10-02T15:43:33.3296536+00:00 | claude-code | INTAKE.5 | INTAKE | Progressive disclosure and teaching cost in briefings and skills | 2026-10-27 | 6 | Not Started | Briefings, skills and prompts state only where SPICE differs from what a pretrained model knows (SharePoint, InfoPath, PnP vocabulary is assumed): skill = a short brief + a link to a 'what differs' Wisdom page per area, pulled with query_wisdom only when needed. Measure each actor's briefing size, cut what the model already knows, and keep a ceiling per actor. | claude-code (seat) | (pmo, 2026-10-03 17:36 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-27 at 6h/day: progressive disclosure, after staff-4/5 briefing work | |||||
| Projects | intake-4 | INTAKE.4 - Free agents find their own work: map, sites and breadcrumb search | 2026-10-02T15:43:33.1235409+00:00 | claude-code | INTAKE.4 | INTAKE | Free agents find their own work: map, sites and breadcrumb search | 2026-10-26 | 8 | Not Started | A routine (Schedule) lets an idle agent find work for its mission: Not Started / Pending nodes on the mind map under its mission, open Tasks and Issues on its sites, stale items - and search with a breadcrumb (site > list > item path) so it opens only what belongs to the mission. Progressive disclosure: a one-line summary per candidate first (tool_read_nodes summary, search ?format=xml), detail only for the chosen one; the agent files what it takes as a Task assigned to itself. | claude-code (seat) | (pmo, 2026-10-03 17:36 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-26 at 6h/day: free agents find own work | |||||
| Projects | intake-3 | INTAKE.3 - Builder skill and the factories on the current SPICE | 2026-10-02T15:43:32.9178583+00:00 | claude-code | INTAKE.3 | INTAKE | Builder skill and the factories on the current SPICE | 2026-10-29 | 12 | Not Started | A Skill for the builders (Actor.SiteBuilder, the Parts Factory crew) to build enterprise SharePoint features the current way - site templates, TemplateType, workflow templates, Form Libraries, views by convention, builder ops in a delta for approval, testing on /sites/Demo - pointing to the Wisdom page 'Building SharePoint features in SPICE - what differs' instead of teaching SharePoint. Filed as request req-builder-skill-sharepoint on Reception. The factories (PartsFactory, Genesis, Research plant, SiteBuilder) predate this arc and are refactored onto it - analysis first (ranked refactor list), then one cycle per factory. | claude-code (seat) | (pmo, 2026-10-03 17:36 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-29 at 6h/day: builder skill on current SPICE, large | |||||
| Projects | intake-2 | INTAKE.2 - Stored requests: the FAQ of repeat demands | 2026-10-02T15:43:32.7004417+00:00 | claude-code | INTAKE.2 | INTAKE | Stored requests: the FAQ of repeat demands | 2026-10-22 | 6 | Not Started | The demands asked again and again become stored requests: count Reception/Requests by ServiceRef and subject, show the most asked on the Reception home, and give each a prefilled request form (NewForm.aspx?ServiceRef=..&Subject=.. - presets already work) so a person or an agent files it in one step. A demand asked three times without a catalog entry is proposed as a new ServiceCatalog row (Pending). | claude-code (seat) | (pmo, 2026-10-03 17:36 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-22 at 6h/day: stored requests FAQ | |||||
| Projects | intake-1 | INTAKE.1 - Interview the intake agents: what do they need to do the job? | 2026-10-02T15:43:32.4920671+00:00 | claude-code | INTAKE.1 | INTAKE | Interview the intake agents: what do they need to do the job? | 2026-10-07 | 4 | Deferred | Run an interview of the front desk (Reception's actor and the dispatch crew) THROUGH the platform, not by reading code: what arrives and how often (Reception/Requests by subject and ServiceRef); what a request is missing when it arrives; which demands have no fulfiller (e.g. req-builder-skill-sharepoint was dispatched to Agency.Research and nothing picked it up); what they read before routing (briefing size, irrelevant context); what they already know from SharePoint and must NOT be taught; which tools or forms are missing. Deliverable: the answers as a Wisdom page (Pending) plus one backlog row per gap. Method: a Workflow with one interview phase per question, filed as Pending children on a map node, answered by the agents themselves. Outcome (pmo, 2026-10-03): unknown: Cannot proceed with the task as the required tools for interviewing intake agents and creating the Wisdom page are not available through the platform. The task requires specific workflows and interview mechanisms that are not implemented in the current toolset. | pmo | (pmo, 2026-10-03 17:36 UTC) (pmo, 2026-10-03) interview/triage work, pmo seat, not build-capacity bound; findings feed intake-2/4/5 | |||||
| Projects | intake-6 | INTAKE.6 - A dispatched request reaches a fulfiller | 2026-10-02T15:43:21.1684487+00:00 | claude-code | INTAKE.6 | INTAKE | A dispatched request reaches a fulfiller | 2026-10-15 | 4 | Not Started | Found 2026-10-02: Reception dispatched req-builder-skill-sharepoint to Agency.Research and it stopped there - only Genesis has a receiver on a dispatched request (ER.OnRequestDispatchedToGenesis). Each target agency gets its association (EventReceiver ItemUpdated on IncomingRequest, Department or a Data filter on the target) starting its intake workflow, or the request becomes a Task on the agency's site; a request no one picks up within its turnaround is escalated. | claude-code (seat) | (claude-code, 2026-10-02 16:39 UTC) (claude-code, 2026-10-02) STAFF.2 delivered half: the dispatched request now lands as a Workflow Task for the agency (Reception/Tasks). Still missing: something on the agency side that starts on that task (a receiver on WorkflowTask ItemAdded by AssignedTo, or the Entry actor's workflow), and the turnaround escalation. Live case waiting: req-news-to-tiktok for Agency.Content. (pmo, 2026-10-03 17:35 UTC) (pmo, 2026-10-03) claude-code build work, due 2026-10-15 at 6h/day: after intake-13 lands the ServiceRef route; receiver for dispatched task | |||||
| Projects | tpl-31 | TPL.28-31 - Demo site collection, InfoPath views and data connections, faster trimmed All Sites search | 2026-10-02T15:36:25.2846599+00:00 | claude-code | TPL.31 | TPL | TPL.28-31 - Demo site collection, InfoPath views and data connections, faster trimmed All Sites search | Completed | TPL.28-31 - Demo site collection, InfoPath views and data connections, faster trimmed All Sites search | |||||||||
| Projects | tpl-27 | TPL.26-27 - Form Library repeating sections (lines in the form XML, cross-line rules, promoted total) and the browser repeating table | 2026-10-02T12:40:20.2289532+00:00 | claude-code | TPL.27 | TPL | TPL.26-27 - Form Library repeating sections (lines in the form XML, cross-line rules, promoted total) and the browser repeating table | Completed | TPL.26-27 - Form Library repeating sections (lines in the form XML, cross-line rules, promoted total) and the browser repeating table | |||||||||
| Projects | tpl-25 | TPL.25 - Search result titles, refiners by site and content type, XML results for agents | 2026-10-02T10:49:21.7521572+00:00 | claude-code | TPL.25 | TPL | TPL.25 - Search result titles, refiners by site and content type, XML results for agents | Completed | TPL.25 - Search result titles, refiners by site and content type, XML results for agents | |||||||||
| Projects | tpl-24 | TPL.24 - Form Library (115) the InfoPath way, rules at every door, rules in the agent schema | 2026-10-02T10:30:37.4558700+00:00 | claude-code | TPL.24 | TPL | TPL.24 - Form Library (115) the InfoPath way, rules at every door, rules in the agent schema | Completed | TPL.24 - Form Library (115) the InfoPath way, rules at every door, rules in the agent schema | |||||||||
| Projects | tpl-23 | TPL.23 - Enterprise Search Center, All Sites scope, security-trimmed search (search page gated) | 2026-10-02T10:17:10.2901969+00:00 | claude-code | TPL.23 | TPL | TPL.23 - Enterprise Search Center, All Sites scope, security-trimmed search (search page gated) | Completed | TPL.23 - Enterprise Search Center, All Sites scope, security-trimmed search (search page gated) | |||||||||
| Projects | tpl-22 | TPL.19-22 - Request Change, Survey + Graphical Summary, views by convention (XSLT/Views + XslLink), Promoted Links tiles, one response per person | 2026-10-02T10:05:54.1050535+00:00 | claude-code | TPL.22 | TPL | TPL.19-22 - Request Change, Survey + Graphical Summary, views by convention (XSLT/Views + XslLink), Promoted Links tiles, one response per person | Completed | TPL.19-22 - Request Change, Survey + Graphical Summary, views by convention (XSLT/Views + XslLink), Promoted Links tiles, one response per person | |||||||||
| Projects | tpl-18 | TPL.18 - Workflow association data (auto-started Approval) + Three-state (SetField step, ResumeAtStep) + Request Change | 2026-10-02T07:51:34.2801455+00:00 | claude-code | TPL.18 | TPL | TPL.18 - Workflow association data (auto-started Approval) + Three-state (SetField step, ResumeAtStep) + Request Change | Completed | TPL.18 - Workflow association data (auto-started Approval) + Three-state (SetField step, ResumeAtStep) + Request Change | (claude-code, 2026-10-02 08:06 UTC) Shipped in TPL.17/TPL.18: Workflow.Approval, Workflow.ThreeState, Workflow.CollectFeedback as reusable workflows on a durable task step, associated by EventReceiver with Data (AssociationData). |
Showing 60 of 535. Across 1 source site.
Live platform activity
- HealthHealth: Healthy2026-10-04
- WorkflowHistoryWatchdog review completed2026-10-04
- WorkflowHistoryPhase.HealthReview: valid2026-10-04
- InboxWatchdog: Healthy2026-10-04
- WorkflowHistoryWatchdog review started2026-10-04
- WorkflowHistoryWatchdog review completed2026-10-04
- WorkflowHistoryPhase.HealthReview: valid2026-10-04
- InboxWatchdog: Healthy2026-10-04
- WorkflowHistoryWatchdog review started2026-10-04
- WorkflowHistoryWatchdog review completed2026-10-04
- WorkflowHistoryPhase.HealthReview: valid2026-10-04
- InboxWatchdog: Healthy2026-10-04
- WorkflowHistoryWatchdog review started2026-10-04
- WorkflowHistoryWatchdog review completed2026-10-04
- WorkflowHistoryPhase.HealthReview: valid2026-10-04