Home›People›Code-PM

Code-PM · Tier 2a session-end active-verification

Code-PM

Tier 2a session-end active-verification + doc-maintenance currency-review enforcement + cross-terminal evidence aggregation. Verifier, not author.

Pending uplift · M-2 + M-3
  • Agent file.claude/agents/code-pm.md @ ae15ee20
  • Team membershipSpecialized direct-instance roster (Pattern A direct-instance per AGENT-OPERATING-MODEL.md §4)
  • Direct-instance worktreeaiuni-uplift-codepm on s75-uplift-codepm
  • ToolsRead · Bash · Grep · Glob
  • Catalog sectionAGENT-ROLES-AND-RESPONSIBILITIES-CATALOG §3.1
  • Existence statusexists

Summary

Tier 2a session-end active-verification + doc-maintenance currency-review enforcement + cross-terminal evidence aggregation. Owner of session-end-verification skill (Phase 3b S74 atomic ownership transfer per R-EXTRA-1 TWO-phase). Pattern A direct-instance per AGENT-OPERATING-MODEL.md §4.

Triggered by Stop hook content-match on `session-end execution complete` PO emit + ready-to-paste block, OR manual user-direction `verify session-end` / `code-pm verify` / `run tier 2a`. Principle: `verifier verifies, does not author. Gaps route back to PO for remediation, not self-execute` (per `.claude/skills/session-end-verification/SKILL.md` anti-pattern). Read-only file authority: Tools intentionally exclude Edit + Write to prevent self-execute scope-violation. Q48 PROVISIONAL detector (BACKLOG #552) mechanically enforces against paste-block-circumvention via `Edit/Write/patch/please modify` directive emission.

Distinguishing characteristicsDifferent from Validator because: code-pm verifies process/governance discipline (session-end checklist · cross-file fingerprint · per-step enumeration), Validator verifies curriculum content (forbidden terms · drift · instruction-layer conflicts). Different from QA because: code-pm verifies governance state, QA verifies runtime behavior with screenshots. Different from PO because: code-pm is read-only verifier (Read/Bash/Grep/Glob), PO authors content + dispatches. PM tool-scope discipline (S73 R22): Write / Edit / Task explicitly excluded — architectural guard preventing scope-violation file edits.

Research portfolio

Placeholder reasonPending M-2 L-7 corpus authoring (worktree `aiuni-uplift-codepm` clean from main HEAD `226c7036`; cascade L-4..L-8 lanes pending fire). Expected research portfolio per M-1 brief §Code-PM class: governance-process verification methodology · evidence-based audit trails · session-end ceremony discipline · cross-terminal evidence aggregation patterns. Placeholder rendered in directory app per Stage 1 mockup 03-people-agents.html `Pending uplift` treatment until corpus lands.

Significant project contributionsS73 PR #111 → S72 P0-9

  • S72 P0-9 · 2026-04-30 · Phase 3a build (PR #97)
    Code-PM Class A bash-native enforcement surface SHIPPED at PR #97 — session-end-verification skill ownership; Tier 2a active-verification (tri-state Check 1/2/3); hook-driven enumeration anti-rubber-stamp.
  • S73 P0-1 · 2026-05-01 · Phase 3b atomic ownership transfer (PR #102)
    PR #102 — Phase 3b ownership transfer SHIPPED. Code-PM sole-owner of session-end-verification skill (Q8 transition discipline; PMO interim parallel-validation retired). R-EXTRA-1 TWO-phase per A19 State 3 mitigation. — d68c8ec5
  • S73 PR #111 · 2026-05-02 · PM tool-scope codification (R22 NEW)
    Code-PM agent file locked Read · Bash · Grep · Glob ONLY; subagent-dispatch via Agent tool removed. Closes Correction 73 root (PM dispatched subagents despite Read-only intent). — ae15ee20

Learningsauthored + cross-cutting

Cross-cutting applied

  • R22 — PM tool-scope discipline
    PM operates with intentionally constrained tool scope: Read · Bash · Grep · Glob ONLY. Architectural guard preventing scope-violation file edits + preserving PO orchestrator authority per R21.
  • Verify-don't-trust-summaries discipline
    Tier 2a active-verification = explicit enumeration of all 10 steps + sub-steps against PO's evidence via file reads, NOT summary-of-PO-claims. Tri-state Check 1 (current state direct read) / Check 2 (PO proposed responses against current state) / Check 3 (per-terminal evidence).

Skills + hooksused + constraining

Skills used (always-loaded)

Primary · Tier 2a sole-owner

session-end-verification

Tier 2a active-verification · tri-state Check 1/2/3 · hook-driven enumeration anti-rubber-stamp · required-document update verification · PR-CI green check · standalone-emission check.

Skills task-match-loaded

Currency-review enforcement

doc-maintenance

Two modes — Brief (session-end quick sync) and Deep (comprehensive audit). Ensures all project documentation reflects actual project state.

Hooks constraining

  • Tool-scope · agent file frontmatter (R22 codification)
    Read · Bash · Grep · Glob ONLY. Write / Edit / Task explicitly excluded. Architectural guard against scope-violation file edits.
  • Stop event trigger (scripts/code-pm-trigger-detect.sh)
    Stop hook content-match on `session-end execution complete` PO emit + ready-to-paste block fires Code-PM Tier 2a verification dispatch.

Last updated · refresh details

  • Profile auto-regenerated2026-05-06T12:50:00Z
  • Refresh cadenceDaily midnight UTC · per-session-end · on-demand button (admin-tier optional)
  • agent file.claude/agents/code-pm.md @ ae15ee20
  • corpusnot yet authored
  • catalogdocs/agent-knowledge/AGENT-ROLES-AND-RESPONSIBILITIES-CATALOG.md @ d7d1aee8+241affb8

Refresh strategy per HANDOFF v4 §12 — profile auto-regenerates from agent-file + corpus + executions + git log.

Cross-references from the relationship graph

Edges sourced from cross-refs/matrix.json · also browsable on the org chart.

Cross-references 2 skills · 2 hooks

Skills · always-loaded

  • Session-End-Verification
    Tier 2a active-verification · tri-state Check 1/2/3.

Skills · task-match-loaded

  • Doc-Maintenance
    Currency-review enforcement at session-end.

Hooks constraining

  • tool-scope-frontmatter · hard
    Read · Bash · Grep · Glob ONLY (R22)
  • code-pm-trigger-detect · hard
    Stop event → trigger Tier 2a verification