ampa-prrd-trdd-kanban

AMPA PRRD / TRDD / Kanban Skill

Overview

This is the MEMBER (programmer)'s role-specific layer of the PRRD / TRDD / Kanban model — the file-based 3-pillars task system (TRDD task documents, PRRD project rules, the design/ kanban). MEMBER is the implementer: when ORCHESTRATOR sets assignee: and moves a TRDD to dev, MEMBER owns that TRDD's movement through dev → testing → ai_review (and back to dev on test failure, up to the project's test-failures threshold). One MEMBER per session; many MEMBERs may work different TRDDs in parallel. The INTEGRATOR — never the MEMBER — owns the final → completed flip after validating the merged PR against the TRDD. For universal mechanics, see the prrd-trdd-kanban skill in ai-maestro-plugin.

This file-based kanban (design/tasks/ + kanban.py) is DISTINCT from the server-side AI Maestro task API (/api/teams/{teamId}/tasks) — that integration is tracked separately (issue #7) and is not required for this skill.

Prerequisites

• The universal prrd-trdd-kanban skill (ai-maestro-plugin — also home of get-prrd.py / prrd-edit.py / findtrdd.py / kanban.py) for shared mechanics and the exempt-operations reference.
• A PRRD at design/requirements/PRRD.md and the 4-zone design/{proposals,tasks,refused,archived}/ folders.
• SERENA MCP for code navigation while implementing.
• A TRDD assigned to this session (assignee: <member-session>, column: dev), received via an AMP notification directly from your ORCHESTRATOR (R6 v3: ORCH ↔ MEMBER is a direct edge; COS guards only the team boundary).

Instructions

1. Read the assigned TRDD frontmatter and body completely.
2. Answer the task-comprehension handshake (loop a) BEFORE coding: send ORCH all five points (restate / files+domains / ambiguities / risks / NPT-EHT) per op-comprehension-handshake (ampa-orchestrator-communication), and WAIT for confirmation.
3. Check relevant-rules: and cite each with get-prrd.py --cite <N>; add any missing rule numbers to relevant-rules:.
4. If delivery: pull-request, create the feature branch git checkout -b feature/TRDD-<id8>-<slug>, then set feature-branch: in the TRDD and bump updated:.
5. Implement to the TRDD's acceptance criteria. Author your DERIVED TASKS (NPT/EHT) directly in design/tasks/ as planned — Tier-0 self-authority covers your own in-scope implementation subtasks. In-dev issues (loop b): raise any ambiguity, blocker, or suspected design flaw to ORCH immediately — NEVER silently improvise around a design flaw (ORCH pulls in ARCH for design, INT for CI/merge).
6. Stage and commit with the TRDD short-ref in the message (git commit -m "feat: <change> (TRDD-<id8>)"), then append the SHA to implementation-commits:.
7. Set column: testing, bump updated:, and run the union of test-requirements: and audit-requirements:; pipe output to reports/member/<TS>-tests-<id8>.log.
8. On pass: set last-test-result: pass, last-test-at: <iso>, column: ai_review. If delivery: pull-request, clear the pre-PR gate (loop c) FIRST: ask ORCH "I believe TRDD- is done — PR now?" per op-pre-pr-gate and open the PR only on the explicit green-light. Then notify ORCH directly.
9. On fail: set last-test-result: fail, increment test-failures:, append a ## Test failure post-mortem <N> section, set column: dev, fix, and re-iterate from step 6.

Approval tiers (what you may do alone)

• Tier 0 (exempt — just do it): dev → testing, testing → ai_review on pass, testing → dev on fail (mechanical bounce), recording implementation-commits:, appending post-mortems, setting feature-branch:, authoring NPT/EHT in design/tasks/, filing PRRD proposals via prrd-edit.py propose.
• Tier 1 (via COS) / Tier 2 (MANAGER via COS): anything beyond your slice — reprioritizing others' work, baseline deviations, cross-team reach, entering the release pipeline. File a proposal in design/proposals/ and route per the approval-tiers rule.
• A MEMBER never self-approves its own releases — USER or MANAGER approve entering the release pipeline, and the INTEGRATOR owns the → completed flip (PRRD S3.1 / S7.1).

Output

• Commits whose messages carry the TRDD short-ref (and PRRD rule when relevant), with SHAs recorded in implementation-commits:.
• Test run logs and last-test-result: / last-test-at: fields.
• Column moves: dev → testing, testing → dev on failure, testing → ai_review on pass — gated by the three dialog loops.

Error Handling

• When test-failures: reaches the project threshold (default 5), stop iterating, write a ## Escalation note in the TRDD body, and AMP-send ORCH directly requesting reassignment or re-design. Wait for ORCH to escalate to ARCH or reassign before resuming.
• When ai_review rejects, treat findings as TRDD updates, implement, re-test, and re-submit. NEVER remove a test that surfaces a real bug — the test is the spec.

Examples

• Assigned a pull-request TRDD: read it, answer the handshake, get ORCH's confirmation, cite PRRD S64.3, branch feature/TRDD-1a2b3c4d-add-cache, implement, commit feat: add request cache (TRDD-1a2b3c4d, PRRD S64.3), record the SHA, move to testing, tests pass, clear the pre-PR gate with ORCH, open the PR, move to ai_review.
• Tests fail at attempt 2: set last-test-result: fail, bump test-failures: to 2, append ## Test failure post-mortem 2, move back to dev, fix the defect, re-commit, re-run.

Resources

For the shared workflow mechanics and the exempt-operations reference, consult the universal prrd-trdd-kanban skill bundled in ai-maestro-plugin; its exempt-operations reference is the authority on which transitions need MANAGER approval. The dialog-loop templates live in ampa-orchestrator-communication (op-comprehension-handshake, op-pre-pr-gate). The MEMBER persona lives in the ai-maestro-programmer-agent-main-agent agent definition.