feat: automatic ACTOR retry on Monitor failure (#92)

* feat: add monitor_failed() for automatic ACTOR retry on Monitor failure

When Monitor returns valid=false, the orchestrator now has a proper
state transition back to ACTOR phase. This fixes the deadlock where
Monitor detects issues but workflow-gate blocks edits during MONITOR.

- Add monitor_failed() for sequential execution mode
- Add wave_monitor_failed() for parallel/wave execution mode
- Both persist feedback to .map/<branch>/monitor_feedback*.md
- Both enforce max_retries with escalation status
- Register as CLI commands: monitor_failed, wave_monitor_failed
- Update map-efficient.md to use orchestrator instead of pseudocode

* chore: formatting cleanup and add DevOpsConf case study link

- Add DevOpsConf 2026 case study link to README.md
- Minor formatting cleanup in ralph-iteration-logger.py, diagnostics.py, map_step_runner.py

* fix: address review findings for monitor_failed

Issues found by Monitor/Predictor/Evaluator review:

- Fix CLI feedback truncation: join all positional args for multi-word
  feedback (was silently dropping words after the first)
- Fix TDD mode: always requeue only ["2.3", "2.4"] on retry, never
  re-run TDD pre-steps (2.25/2.26) that were already completed
- Fix feedback overwrite: use retry-numbered filenames
  (monitor_feedback_retry{N}.md) per error-patterns.md lesson
- Extract _write_feedback_file() helper to DRY the feedback persistence
- Update Phase D wave retry pseudocode to use wave_monitor_failed
- Clarify Actor feedback file path comment (explicit, not "automatic")
- Add 17 unit tests: TestMonitorFailed (10) + TestWaveMonitorFailed (7)

* fix: address review findings and add reopen_for_fixes for post-review edits

Review findings (11 items):
- Extract _check_retry_limit() helper to DRY up monitor_failed/wave_monitor_failed
- Add phase guard to monitor_failed() rejecting non-MONITOR phase calls
- Add --feedback CLI flag for symmetric argument passing
- Fix SUBTASK_ID placeholder, feedback filename comment, Continue label text
- Document serial/wave retry counter invariant in StepState
- Update ARCHITECTURE.md CLI command list

New feature: reopen_for_fixes()
- Transitions COMPLETE → ACTOR so workflow gate unlocks for review fixes
- Solves: /map-review finds issues but edits blocked in COMPLETE phase
- map-review.md updated with "Workflow Gate Unlock" section

Tests: 13 new tests (6 reopen_for_fixes + 4 review findings + 3 integration)

* fix: add test quality gate to map-tdd to prevent trivial "2+2=4" tests

TEST_WRITER prompt now includes explicit anti-patterns:
- Every test must verify semantic behavior, not just single branch execution
- Tests must assert multiple consequences (side effects, return values, state)
- Prefer scenario-based tests over single-field unit checks
- Aim for 60%+ semantic tests

TEST_FAIL_GATE now includes quality gate:
- Classifies tests as semantic vs trivial
- Rejects if >40% tests are trivial single-branch checks
- Sends feedback to TEST_WRITER to merge trivial tests into richer scenarios

Addresses feedback: 3/9 tests were trivial branch checks in production use.

* fix: prevent stale Red-phase comments in TDD test files

TEST_WRITER could add temporal comments like "currently FAILS" or
"expected to FAIL until fix is applied" that become stale after the
Green phase. Actor in code_only mode cannot clean them up.

- Add rule to TEST_WRITER prohibiting temporal state-marking comments
  (map-tdd.md, map-efficient.md, actor.md)
- Add TDD Refactor cleanup step after Actor returns, before Monitor,
  to scan and clean any stale Red-phase markers in test files

* chore: remove deprecated skills (map-workflows-guide, map-cli-reference, map-debate)

Remove unused skill definitions and their trigger rules from both
dev config (.claude/) and shipped templates (src/mapify_cli/templates/).

* fix: remove map-debate from expected command templates in tests

* chore: remove all map-debate references from docs, tests, and plugin metadata

Clean up remaining mentions after command/skill deletion.
CHANGELOG entries preserved as historical record.

* chore: ignore .map/scripts/.map/ in gitignore

Author: azalio

.claude-plugin/PLUGIN.md

@@ -39,7 +39,6 @@ MAP (Modular Agentic Planner) is a cognitive architecture that orchestrates 11 s
 - `/map-efficient` — implement features, refactor code, complex tasks with full MAP workflow
 - `/map-debug` — debug issues using MAP analysis
 - `/map-fast` — small, low-risk changes with minimal overhead
-- `/map-debate` — multi-variant synthesis with Opus arbiter
 - `/map-review` — comprehensive review of changes
 - `/map-check` — quality gates and verification
 - `/map-plan` — architecture decomposition

.claude-plugin/marketplace.json

@@ -33,7 +33,7 @@
       "features": [
         "11 specialized agents (TaskDecomposer, Actor, Monitor, Predictor, Evaluator, Reflector, DocumentationReviewer, Debate-Arbiter, Synthesizer, Research-Agent, Final-Verifier)",
         "5 Claude Code hooks for automation",
-        "12 slash commands (/map-efficient, /map-debug, /map-fast, /map-debate, /map-review, /map-check, /map-plan, /map-task, /map-tdd, /map-release, /map-resume, /map-learn)",
+        "11 slash commands (/map-efficient, /map-debug, /map-fast, /map-review, /map-check, /map-plan, /map-task, /map-tdd, /map-release, /map-resume, /map-learn)",
         "Professional code review integration",
         "Cost optimization (40-60% savings)"
       ],

.claude-plugin/plugin.json

@@ -31,7 +31,7 @@
   "features": [
     "11 specialized MAP agents (TaskDecomposer, Actor, Monitor, Predictor, Evaluator, Reflector, DocumentationReviewer, Debate-Arbiter, Synthesizer, Research-Agent, Final-Verifier)",
     "5 Claude Code hooks for automation (validate-agent-templates, auto-store-knowledge, enrich-context, session-init, track-metrics)",
-    "12 slash commands (/map-efficient, /map-debug, /map-fast, /map-debate, /map-review, /map-check, /map-plan, /map-task, /map-tdd, /map-release, /map-resume, /map-learn)",
+    "11 slash commands (/map-efficient, /map-debug, /map-fast, /map-review, /map-check, /map-plan, /map-task, /map-tdd, /map-release, /map-resume, /map-learn)",
     "Chain-of-thought reasoning with sequential-thinking MCP",
     "Semantic pattern search with embeddings cache",
     "Cost optimization with per-agent model selection (haiku/sonnet/opus)"

.claude/agents/actor.md

@@ -276,6 +276,9 @@ Rules:
 5. Include edge cases from the spec's `## Edge Cases` section if available in the packet.
 6. Use standard test patterns for the project's language and framework.
 7. Tests SHOULD fail when run (implementation doesn't exist yet). This is expected.
+8. Do NOT add temporal comments about test failure status (e.g., "currently FAILS",
+   "expected to FAIL", "will PASS once fix is applied"). Write tests as permanent,
+   clean code — the Red/Green state is transient and must not leak into comments.
 
 Output:
 - Test files created via Write tool