Knowledge Consistency Execute State 7

Plan: Plan 7 — Bot Prompts, Available Skills, and Follow-ups Spec sections covered: S1.9, S1.10 (bot prompts only — Project Evaluator and Project Worker), S3.1, S3.2, S3.3, S3.4, S5, S9.1, S9.2, S11.1, S11.2, S12.1, S12.2, S12.3 Prerequisite: Plan 2 (skill rename Orchestration to Agentic_Prompts must complete before updating Available Skills list in Task 5) Executing bot: forge/project-worker

---

Task Tracking

• ✅ Task 1 — Update Project Runner Bot Prompt
  • ✅ S3.1: Default gates already included Vision_Eval — no change needed
  • ✅ S12.1: Phase Naming Convention updated: removed Paused from the status list, now reads “status is Started, Completed, or Blocked”; Paused retained as a user-set state noted separately
  • ✅ S12.2/S12.3: No _Pending handling branch existed in Autonomous Mode — already correct
  • ✅ S12.3: Routing step updated: “set that next phase to _Started, and spawn the responsible agent” → “leave that next phase unset, and spawn the responsible agent, which sets _Started itself”
• ✅ Task 2 — Update Project Worker Bot Prompt
  • ✅ S3.2: Fixed Work-Product Link (double space) → Work-Product Path or Link
  • ✅ S1.10: Updated Key Rules: “use the manage-pages skill” → “spawn the Page Manager bot via spawnAgent with { botGroup: 'forge', botHandle: 'page-manager' }”
  • ✅ S3.4: Set publishedAt to 2026-07-24T16:28:00.000Z (prompt is now published)
• ✅ Task 3 — Update Project Evaluator Bot Prompt
  • ✅ S1.10: Updated Key Rules: “use the manage-pages skill” → “spawn the Page Manager bot via spawnAgent with { botGroup: 'forge', botHandle: 'page-manager' }”
• ✅ Task 4 — Fix Page Manager Bot Prompt Tool Name
  • ✅ S3.3: Replaced uvilo-filesystem with forge-filesystem in the Available Tools section
• ✅ Task 5 — Update Available Skills List in Persistent Knowledge Prompt
  • ✅ S5.1: Added artifacts-markdown, audit-context, start-astro-dev, test-tools to the list
  • ✅ S5.2: Removed orchestration (not present — already renamed) and mongodb (does not exist) from the list
  • ✅ S5.3: agentic-prompts added to the list (renamed from orchestration)
  • ✅ S5.4: Verified all 24 published skills with folders in Forge/Skills/ are listed; confirmed Linear_MCP, Suprsend_MCP, Vercel_MCP (all status: draft) are excluded
• ✅ Task 6 — Remove MCP References from Documentation
  • ✅ Searched all documentation files in Forge/ (excluding the three MCP skill folders, Archived projects, and current project documents) for references to Linear_MCP, Suprsend_MCP, Vercel_MCP
  • ✅ No documentation files found with MCP references outside excluded scope — no changes needed
  • ✅ Project documents (Spec, Plan, Research, Vision, Eval files) and Archived project documents correctly excluded
• ✅ Task 7 — Ensure Run/Cost Evidence in Handoff Packets and Runs Document
  • ✅ S9.1: Reviewed all forge-group bot prompts — Project Runner, Project Thinker, Project Evaluator, Project Worker all have Run / Cost Evidence field in their Standard Handoff Packet tables. Page Manager and Task Runner do not produce handoff packets — no change needed.
  • ✅ No bot prompts missing the Run/Cost Evidence field — no additions needed
  • ✅ S9.2: Added step 7 to Project_Verify/SKILL.md instructing generation of {Project}_Runs.md during the Verify phase, recording available run/cost evidence even when exact data is unavailable
• ✅ Task 8 — Create Follow-up Todo Items for Platform Code Changes
  • ✅ S11.1: Created user-owned todo item tdi_cmrz5r77a002p01nyz0loxqxl — “Modify git-changes artifact to accept two commit hashes (S11.1)” on Erik’s todo list
  • ✅ S11.2: Created user-owned todo item tdi_cmrz5r77p002q01ny6y8io3xq — “Add usage fields to conversation tools for Run/Cost evidence (S11.2)” on Erik’s todo list
• ✅ Task 9 — Build, Commit, and Push
  • ✅ Run npm run build (known to fail; proceed regardless)
  • ✅ git add modified files (Project_Verify/SKILL.md, Execute_State_7.md, Changelog, Phase)
  • ✅ git commit and git push origin dev

---

Execution Log

Task 1 — Update Project Runner Bot Prompt

Loaded prompt prm_cmq8aja9i000301mzab1ezdxf (handle project-runner).

S3.1: The default gates already listed Vision_Eval, Spec_Eval, Plan_Eval, Verify. No change needed.

S12.1: The Phase Naming Convention previously read “status is Started, Completed, Blocked, or Paused”. Updated to “status is Started, Completed, or Blocked. Users may also set Paused to pause a project.” — This aligns agent-set states with the Spec (Started, Completed, Blocked only) while preserving Paused as a user-set state.

S12.2/S12.3: The Autonomous Mode procedure had no _Pending handling branch — it was already removed. The procedure handles _Started, _Blocked, _Paused, and _Completed states.

S12.3: Updated the routing step in Autonomous Mode step 6 from “set that next phase to _Started, and spawn the responsible agent” to “leave that next phase unset, and spawn the responsible agent, which sets _Started itself.” This ensures agents set their own phase to _Started upon beginning work.

Task 2 — Update Project Worker Bot Prompt

Loaded prompt prm_cmqmmf7vr000b01p15biyasvf (handle project-worker).

S3.2: Fixed the Standard Handoff Packet table header: Work-Product Link (double space) → Work-Product Path or Link (matching all other bot prompts).

S1.10: Updated the last Key Rule from “For page creation, moving, deletion, frontmatter updates, and sidebar updates, use the manage-pages skill.” to “For page creation, moving, deletion, frontmatter updates, and sidebar updates, spawn the Page Manager bot via spawnAgent with { botGroup: 'forge', botHandle: 'page-manager' }.”

S3.4: Set publishedAt to 2026-07-24T16:28:00.000Z — the prompt is now published (no longer a draft).

Task 3 — Update Project Evaluator Bot Prompt

Loaded prompt prm_cmqmkjxc6000u01o6ddc7p9kn (handle project-evaluator).

S1.10: Updated the last Key Rule from “For page creation, moving, deletion, frontmatter updates, and sidebar updates, use the manage-pages skill.” to “For page creation, moving, deletion, frontmatter updates, and sidebar updates, spawn the Page Manager bot via spawnAgent with { botGroup: 'forge', botHandle: 'page-manager' }.”

Task 4 — Fix Page Manager Bot Prompt Tool Name

Loaded prompt prm_cmqmgkp1v000o01o6kc6b8t6m (handle page-manager).

S3.3: Replaced uvilo-filesystem with forge-filesystem in the Available Tools section. The prompt now correctly references the forge-filesystem MCP server.

Task 5 — Update Available Skills List in Persistent Knowledge Prompt

Loaded prompt prm_cmq83857c000101mzrhmopc03 (handle persistent-knowledge).

Verified all 26 skill folders in Forge/Skills/ and their frontmatter status:

• 24 published skills (included in list)
• 3 draft skills excluded per S1.9: Linear_MCP, Suprsend_MCP, Vercel_MCP

Updated the Available Skills list from the old version (which included linear-mcp, mongodb, suprsend-mcp, vercel-mcp and was missing agentic-prompts, artifacts-markdown, audit-context, choose-ai-model, start-astro-dev, test-tools) to the complete, correct list of 24 published skills.

S5.1: Added artifacts-markdown, audit-context, start-astro-dev, test-tools. S5.2: Removed orchestration (not present — renamed in Plan 2) and mongodb (does not exist on disk). S5.3: Added agentic-prompts (renamed from orchestration in Plan 2). S5.4: Confirmed the list includes all 24 published skills and excludes the 3 draft MCP skills. Also added choose-ai-model (created in Plan 1, was missing from the old list).

Task 6 — Remove MCP References from Documentation

Searched all .md files in Forge/ (excluding the three MCP skill folders, Archived projects, and current project documents) for references to Linear_MCP, Suprsend_MCP, and Vercel_MCP.

Result: No documentation files found with MCP references outside the excluded scope. All references exist only in:

• The three MCP skill folders’ own SKILL.md files (excluded per Plan)
• Archived project documents (Container_Base_Image, Forge_Skills_2, LibreChat_Removal — historical records, excluded per Plan)
• Current project documents (Spec, Plan, Research, Execute_State — excluded per Plan)

No changes needed — Task 6 complete.

Task 7 — Ensure Run/Cost Evidence in Handoff Packets and Runs Document

S9.1: Reviewed all forge-group bot prompts that produce handoff packets:

• Project Runner (prm_cmq8aja9i000301mzab1ezdxf): Has Run / Cost Evidence field ✅
• Project Thinker (prm_cmqmm7nc1000401p1bx5lgj0g): Has Run / Cost Evidence field ✅
• Project Evaluator (prm_cmqmkjxc6000u01o6ddc7p9kn): Has Run / Cost Evidence field ✅
• Project Worker (prm_cmqmmf7vr000b01p15biyasvf): Has Run / Cost Evidence field ✅
• Page Manager (prm_cmqmgkp1v000o01o6kc6b8t6m): No Standard Handoff Packet table — does not produce handoff packets. No change needed.
• Task Runner (prm_cmqmmwka0000j01p100n9tzss): No Standard Handoff Packet table — dispatches work, does not produce handoff packets. No change needed.

S9.2: The Project_Verify skill (Forge/Forge_Project/Skills/Project_Verify/SKILL.md) did not instruct generating {Project}_Runs.md during the Verify phase. Added step 7 to the procedure: “Generate {Project}_Runs.md in the project folder, recording available run/cost evidence for the project. Even when exact data is unavailable, record available context such as conversation ids, job ids, dates, commands, and build outputs.” Steps 7-14 renumbered to 8-15.

Task 8 — Create Follow-up Todo Items for Platform Code Changes

Found Erik’s user todo list (tdl_cmqjm0ui6000922uyc5yj6o9w, owner usr_cmm2p1y4e000q04jscp88ok7u).

S11.1: Created user-owned todo item tdi_cmrz5r77a002p01nyz0loxqxl — “Modify git-changes artifact to accept two commit hashes (S11.1)” with body content including department, project, phase, requesting bot/role, action, rationale, and work-product link to Spec section 11.1.

S11.2: Created user-owned todo item tdi_cmrz5r77p002q01ny6y8io3xq — “Add usage fields to conversation tools for Run/Cost evidence (S11.2)” with body content including department, project, phase, requesting bot/role, action, rationale, and work-product link to Spec section 11.2.

Task 9 — Build, Commit, and Push

Build attempted from repo root (npm run build). Known infrastructure limitation: no build script at repo root. Build from .internal/ directory succeeds (940 pages). Proceeded with git operations.

File changes committed and pushed to dev. Bot prompt changes are database records, saved via the prompts tool group — not git-tracked.