Skip to content

feat: GH-issue cluster #17-#23 work-down + accumulated cleanup#24

Merged
projectious merged 2 commits into
mainfrom
feat/gh-issue-cluster-2026-05-09
May 9, 2026
Merged

feat: GH-issue cluster #17-#23 work-down + accumulated cleanup#24
projectious merged 2 commits into
mainfrom
feat/gh-issue-cluster-2026-05-09

Conversation

@projectious
Copy link
Copy Markdown
Contributor

Summary

Closes/addresses 7 GitHub issues in 4 sequenced waves, plus accumulated working-tree state from prior sessions (8 applied migrations 0.23.11→0.25.6, legacy .agents/skills/pk-* and .claude/commands/pk-*.md shim removal, bulk skill-catalog stub removal under context/skills/{data-ai,design,devops,documents,engineering,product}/).

7 of 7 GH-issue WorkItems are in review state apart from VastVale (#20) which is in-progress as a decomposed epic with 5 sub-WorkItems (4 backlog, 1 review).

Wave 1 — pk-doctor SKILL.md docs (closes #23)

Documented 6 previously-undocumented check modules in context/skills/processkit/pk-doctor/SKILL.md:

  • v2_contracts, context_hygiene, schema_vocabulary, migration_integrity, mcp_gateway, skill_dag

Wave 2 — v1-drift cluster (closes #21, #22)

  • New pk-doctor check v1_entity_drift (~265 lines): per-file v1 frontmatter scan with v2 successor table.
  • Router & skill-finder & index-management v1-entity penalty (default 0.3, configurable via task-router/mcp/user_config.json) — applied at SKILL.md apiVersion level AND at index-level entity reads (query_entities, get_entity, search_entities).

Wave 3 — agent-dispatch cluster (closes #17, #18, #19)

  • compliance-contract.md rebuilt: 6 sections (On session start / Sub-agent dispatch / Tool routing / Entity writes / Decisions / Prohibitions), action-first phrasing, ~10 positive : 4 negative imperatives.
  • New sub-agent-dispatch clause: route_task → use recommended_team_member_slug + recommended_model_class.
  • task-router response shape extended: recommended_team_member_slug (resolves Cora for 5/14 PM-owned domain groups today; engineering-role mapping is a follow-up), recommended_model_class.
  • AGENTS.md "Before sub-agent dispatch" section added.
  • New pk-doctor check team_member_exports: warns when active TeamMembers lack .claude/agents/<slug>.md exports (3 WARNs against this repo).
  • Slim per-turn hook injection via <!-- BEGIN HOOK --> / <!-- END HOOK --> markers in compliance-contract.md; emit_compliance_contract.py extracts the slim block on UserPromptSubmit (~14 lines) and the full contract on SessionStart. Per-turn token tax cut from ~60 to ~14 lines. (Per-turn hook → session-skill migration, gh#19 rec 1, was rejected — see DEC-LivelyClover.)
  • export_claude_subagent now writes a self-describing TeamMember + RuntimeBinding header in generated agent files.
  • 4 large infrequent skills (skill-builder, skill-reviewer, team-creator, agent-management) marked skillOverrides=name-only in the documented harness settings.
  • New docs/harness-claude-code.md + docs/harness-claude-code.settings.example.json.

Wave 4 — team-creator v2 design epic (in-progress, #20)

  • VastVale split into 5 sub-WorkItems: TidyAsh (RoleSlot primitive), LuckyWren (catalog-driven), RapidLily (consultant type), SwiftReef (budget projection), MerryPlum (codename rename).
  • Cross-gap design synthesis: ART-20260509_1836-SmartPanda — recommends RoleSlot+TeamMember hybrid identity model, direct catalog consumption, consultant TeamMember type with engagement_window, budget_projection in chartering DEC inputs_snapshot, two-phase additive-then-cutover migration.
  • 4 architectural decisions accepted (DEC-CoolBadger): RoleSlot lives in team-manager (no skill inflation), v0.16.0 capacity fields removed cleanly, consultants use the same TeamMember schema, live get_pricing for budget unit costs.
  • SUB-5 codename rename SHIPPED: 12 replacements across team-creator/ tree.

Decisions recorded (5)

ID Subject
DEC-SmoothLynx Wave 1 dispatch approval
DEC-LivelyClover DaringRaven rec 1 reject; rec 4 accept (slim hook, no migration)
DEC-MerryFjord WarmOak entity-level extension
DEC-MightySky VastVale v2-in-place + dispatch architect
DEC-CoolBadger 4 design Q&A answers

Tests

  • task-router: 9 tests passing (added v1 penalty + sub-agent dispatch field tests)
  • skill-gate: 87 hook tests passing (added slim-payload + full-contract assertions)
  • team-manager: 56 tests passing (added export_claude_subagent header assertions)
  • pk-doctor v1_entity_drift: smoke-tested via py_compile only (sandbox blocked pyyaml import)

Test plan

  • Restart MCP servers; verify route_task returns recommended_team_member_slug / recommended_model_class
  • Run pk-doctor; verify new checks (v1_entity_drift, team_member_exports) emit expected findings
  • Verify slim hook payload appears on UserPromptSubmit, full contract on SessionStart
  • Confirm mirror parity: diff -r --exclude=__pycache__ context/skills/processkit/ src/context/skills/processkit/

Open follow-ups (not in this PR)

  • WildPanda P2: recommended_team_member_slug resolves only 5/14 PM-owned groups; add ROLE-software-engineer mapping.
  • WarmOak: semantic_search_entities and hybrid_search_entities not yet covered by v1 penalty.
  • VastVale Q4 (auto-deactivate consultants on Scope close), Q6 (ROLE-assistant archetype mapping target) — defaulting to architect's recommendations.
  • Live .claude/settings.json is _processkit_managed (installer-rewritten) — skillOverrides and env.ENABLE_TOOL_SEARCH=auto documented but not auto-applied.
  • TidyAsh (SUB-1, RoleSlot primitive) is up next.

🤖 Generated with Claude Code

claude added 2 commits May 9, 2026 23:23
Wave 1 (#23): document 6 pk-doctor check modules (v2_contracts,
context_hygiene, schema_vocabulary, migration_integrity, mcp_gateway,
skill_dag).

Wave 2 (#21, #22): v1 entity penalty in router/skill-finder/index-
management with configurable v1_entity_penalty knob (default 0.3); new
pk-doctor v1_entity_drift check.

Wave 3 (#17, #18, #19): compliance-contract.md rebuild (6 sections,
positive-leaning, action-first); sub-agent-dispatch clause; task-router
recommended_team_member_slug + recommended_model_class fields; AGENTS.md
"Before sub-agent dispatch" section; new pk-doctor team_member_exports
check; slim hook injection (~14 lines) via BEGIN/END HOOK markers;
export_claude_subagent self-describing TeamMember + RuntimeBinding
header; harness docs.

Wave 4 (#20): team-creator v2 design epic split into 5 sub-WorkItems
(TidyAsh/LuckyWren/RapidLily/SwiftReef/MerryPlum); cross-gap design
synthesis filed as ART-20260509_1836-SmartPanda; SUB-5 codename rename
shipped (12 replacements).

5 cross-cutting decisions recorded (SmoothLynx, LivelyClover,
MerryFjord, MightySky, CoolBadger).

Includes accumulated working-tree state from prior sessions: applied
8 migrations 0.23.11 -> 0.25.6, dropped legacy .agents/skills/pk-* and
.claude/commands/pk-*.md shims, removed bulk skill catalog stubs.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Foundational RoleSlot primitive WorkItem moves backlog -> in-progress
ahead of dispatch on a separate branch.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

2 participants