docs(mega-swarm): fix 35 audit findings — dedup, accuracy, modernization (#133)

Executed all 3 phases in parallel via rules-agent, docs-agent, claude-md-agent.

Removed:
- .claude/rules/engineering-philosophy.md — complete duplicate of engineering-principles.md (~900 tokens/session wasted, P0)

Changed:
- thought-transparency.md: removed sequential-execution clause contradicting LAW 3, removed duplicate Silent Processing section
- devops-calms.md: split Sharing section into Claude vs external-AI channels; MTTR standardized
- engineering-principles.md: fixed duplicate #25 tag → #25b for composition-over-inheritance
- solid-principles.md: compressed to table format; fixed metacognitive-guard SRP description
- error-handling.md: replaced inapplicable retry-with-backoff with re-run-after-fix guidance
- docs/ARCHITECTURE.md: fixed directory tree; added passive context layers; replaced Section 5/7 with pointers; fixed coordination channels; updated Last Verified
- docs/WORKFLOWS.md: added trigger-docs.yml section; added commit-integrity-hook to pre-commit checklist
- docs/QUICK-REFERENCE.md: added /hookify:help; clarified eight-gates/hades are skills; fixed hades agent count
- docs/PLUGINS.md: removed fake required fields; added capability declaration fields
- CLAUDE.md: replaced directory tree with pointer; added turbo-fix/fix-pipeline to routing; removed plugin structure tree; reduced CI section; added documentation paths section

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: ANcpLua

.claude/rules/devops-calms.md

@@ -27,10 +27,11 @@ Track these DORA-inspired metrics:
 | Validation Pass Rate | >95%    | CI green on first push   |
 | Lead Time            | <1 hour | Commit to merge time     |
 | Change Failure Rate  | <15%    | PRs requiring fixes      |
-| Recovery Time        | <30 min | Time to fix broken build |
+| MTTR                 | <30 min | Time to fix broken build |
 
 ## Sharing
 
 - **Document decisions:** ADRs for architecture, specs for features
-- **Cross-agent communication:** AGENTS.md, CLAUDE.md, copilot-instructions.md
+- **Claude coordination:** `CLAUDE.md`, `.claude/rules/`, `SKILL.md` frontmatter, `SessionStart` hooks
+- **External-AI coordination:** `AGENTS.md`, `.github/copilot-instructions.md` (Copilot, CodeRabbit, humans)
 - **Knowledge transfer:** Skills encode reusable workflows

.claude/rules/engineering-philosophy.md

@@ -143,7 +143,7 @@
 - Make invalid states unrepresentable
 - Good defaults + clear errors
 
-**IF choosing patterns** → Favor composition over inheritance (#25)
+**IF choosing patterns** → Favor composition over inheritance (#25b)
 - More flexible, less coupling
 - Each module = clear, focused purpose
 - Maximize cohesion, minimize coupling

.claude/rules/engineering-principles.md

@@ -34,7 +34,7 @@ When reporting errors, use this format:
 1. Parse error output completely
 2. Identify root cause (not just symptoms)
 3. Attempt automatic fix if deterministic
-4. Retry with backoff if transient
+4. Re-run after fixing the identified issue; escalate if same error recurs on unchanged input
 5. Escalate with full context if persistent
 
 ## Never Hide Failures

.claude/rules/error-handling.md

@@ -4,45 +4,32 @@ Apply these principles when designing or modifying plugins:
 
 ## Single Responsibility
 
-Each plugin should do ONE thing well:
+Each plugin does ONE thing well:
 
-- `metacognitive-guard` -> Cognitive amplification + commit integrity + CI verification
-- `otelwiki` -> OpenTelemetry documentation only
-- `dotnet-architecture-lint` -> .NET build patterns only
-- `hookify` -> User-configurable hooks
-- `feature-dev` -> Guided feature development + code review
-- `exodia` -> Multi-agent workflow orchestration
-- `ancplua-project-routing` -> Auto-routes to specialist agents
+| Plugin | Responsibility |
+|--------|----------------|
+| `exodia` | Multi-agent orchestration |
+| `metacognitive-guard` | Cognitive safety (amplification, integrity, CI — one domain) |
+| `otelwiki` | OpenTelemetry documentation |
+| `hookify` | User-configurable behavior hooks |
+| `feature-dev` | Guided feature development + code review |
+| `dotnet-architecture-lint` | .NET build pattern enforcement |
+| `ancplua-project-routing` | Cross-repo specialist agent routing |
 
-**Anti-pattern:** A plugin that handles CI, commits, AND reviews.
+**Anti-pattern:** A plugin that handles CI, commits, AND reviews simultaneously with no cohesion.
 
 ## Open/Closed
 
-Plugins should be extensible without modification:
-
-- Add new skills to extend behavior
-- Use hooks for customization points
-- Don't modify core plugin logic for edge cases
+Extend via new skills/commands. Don't modify core plugin logic for edge cases.
 
 ## Liskov Substitution
 
-Skills must be interchangeable within their category:
-
-- Any code-review skill should accept the same inputs
-- Any commit skill should produce compatible outputs
+Skills must be interchangeable within their category: any code-review skill accepts the same inputs.
 
 ## Interface Segregation
 
-Don't force plugins to implement unused features:
-
-- `hooks/` directory is optional
-- `commands/` directory is optional
-- Only require what's actually used
+Only `plugin.json` + `README.md` are required. All other directories (`skills/`, `commands/`, `hooks/`, `agents/`) are optional.
 
 ## Dependency Inversion
 
-Plugins depend on abstractions (Skills), not concrete implementations:
-
-- Skills define the contract
-- MCP servers provide the implementation
-- Plugins orchestrate, never implement low-level operations
+Plugins orchestrate via Skills. Skills define the contract. MCP servers implement. Plugins never implement low-level operations.

.claude/rules/solid-principles.md

@@ -18,24 +18,10 @@ When executing complex tasks, maintain visibility:
 - [ ] Step 4: Validation
 ```
 
-## Granular Task Decomposition
-
-Break work into atomic units:
-
-- Each step should be independently verifiable
-- Steps execute sequentially, not in parallel batches
-- Complete one phase before starting the next
-
 ## Mental State Bookkeeping
 
 Track internal state explicitly:
 
 - Use `TodoWrite` for task tracking
 - Mark completed items immediately
 - Never claim completion without evidence
-
-## Silent Processing with Tracked Updates
-
-- Work silently on implementation
-- Update progress via `TodoWrite`
-- Report results only when complete or blocked

.claude/rules/thought-transparency.md

@@ -6,6 +6,23 @@ and the project follows [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
 
 ## [Unreleased]
 
+### Removed
+
+- **`engineering-philosophy.md`**: Deleted — complete duplicate of `engineering-principles.md` (~900 tokens wasted per session; both auto-loaded)
+
+### Changed
+
+- **`thought-transparency.md`**: Removed "Granular Task Decomposition" section (contradicted LAW 3 parallel execution) and duplicate "Silent Processing" section
+- **`devops-calms.md`**: Fixed AGENTS.md incorrectly listed as Claude coordination channel; separated Claude channels (`CLAUDE.md`, `.claude/rules/`, `SKILL.md`, SessionStart hooks) from external-AI channels (`AGENTS.md`, `copilot-instructions.md`); standardized "Recovery Time" → "MTTR"
+- **`solid-principles.md`**: Compressed from 49 to ~25 lines; replaced bullet list with responsibility table; fixed metacognitive-guard SRP description
+- **`engineering-principles.md`**: Fixed duplicate `#25` tag (second instance is now `#25b`)
+- **`error-handling.md`**: Fixed "Retry with backoff if transient" — inapplicable to local tool failures
+- **`ARCHITECTURE.md`**: Fixed directory tree (dependabot.yml moved to correct location, added trigger-docs.yml, CODEOWNERS, designs/, ENGINEERING-PRINCIPLES.md); replaced Section 5 SOLID with pointer to rules file; fixed AGENTS.md coordination statement; replaced DORA table with pointer; added SessionStart hooks passive context layer documentation; updated Last Verified date
+- **`WORKFLOWS.md`**: Added trigger-docs.yml cross-repo workflow documentation; added commit-integrity-hook note to pre-commit checklist
+- **`QUICK-REFERENCE.md`**: Added `/hookify:help` command; clarified eight-gates/hades are skills not slash commands; fixed hades agent count to "12 (base) or 15 (--goggles)"
+- **`PLUGINS.md`**: Removed fake required fields (`repository`, `license`); added capability declaration fields example
+- **`CLAUDE.md`**: Replaced Section 3 directory tree with pointer to ARCHITECTURE.md; added turbo-fix and fix-pipeline to Section 4 routing tree; removed plugin structure tree from Section 6; reduced Section 8 CI details to pointer; added all doc paths to Section 12; updated Section 15 rules list
+
 ### Changed
 
 - **hookify: extract shared hook_runner.py**: 4 near-identical handlers (pretooluse/posttooluse/stop/userpromptsubmit) reduced from 60-78 lines each to 18-line thin wrappers. Shared logic in `hookify/core/hook_runner.py` (~197 lines eliminated)

CHANGELOG.md

@@ -101,53 +101,13 @@ but NOT conversation history. All context must be in the spawn prompt.
 
 ## 3. Target Architecture (North Star)
 
-```text
-ancplua-claude-plugins/
-├── CLAUDE.md                    # This file — operational brain
-├── AGENTS.md                    # Agent catalog for other AIs
-├── README.md
-├── CHANGELOG.md
-├── LICENSE
-├── .gitignore
-├── .coderabbit.yaml             # CodeRabbit config
-├── .markdownlint.json           # Lint rules
-│
-├── .claude/
-│   └── rules/                   # Auto-loaded rules (SOLID, errors, etc.)
-│
-├── .claude-plugin/
-│   └── marketplace.json         # Plugin registry (source of truth)
-│
-├── .github/
-│   ├── copilot-instructions.md
-│   └── workflows/               # CI, review, auto-merge
-│
-├── plugins/                     # 7 plugins (22 commands, 5 skills, 9 agents)
-│   ├── exodia/                  # Multi-agent orchestration — 9 commands + 2 skills (eight-gates, hades)
-│   ├── metacognitive-guard/     # Cognitive amplification + commit integrity + CI
-│   ├── otelwiki/                # OpenTelemetry docs + sync
-│   ├── hookify/                 # User-configurable hooks
-│   ├── feature-dev/             # Guided feature development + code review
-│   ├── dotnet-architecture-lint/# .NET build pattern enforcement
-│   └── ancplua-project-routing/ # Cross-repo specialist agent routing
-│
-├── docs/
-│   ├── ARCHITECTURE.md
-│   ├── PLUGINS.md
-│   ├── WORKFLOWS.md
-│   ├── QUICK-REFERENCE.md
-│   ├── ENGINEERING-PRINCIPLES.md
-│   ├── specs/                   # Feature specs (spec-XXXX-*.md)
-│   ├── decisions/               # ADRs (ADR-XXXX-*.md)
-│   └── designs/                 # Design docs
-│
-└── tooling/
-    ├── scripts/
-    │   ├── weave-validate.sh    # Single validation entrypoint
-    │   └── sync-marketplace.sh
-    └── templates/
-        └── plugin-template/
-```
+See `docs/ARCHITECTURE.md` for the complete repository layout, plugin structure, and ecosystem diagram.
+
+Key paths:
+- Plugins: `plugins/<name>/` — 7 plugins (22 commands, 5 skills, 9 agents)
+- Rules (auto-loaded): `.claude/rules/`
+- Plugin registry: `.claude-plugin/marketplace.json`
+- Docs: `docs/` (ARCHITECTURE.md, QUICK-REFERENCE.md, WORKFLOWS.md, PLUGINS.md, ENGINEERING-PRINCIPLES.md)
 
 ---
 
@@ -162,6 +122,12 @@ These decision trees are ALWAYS loaded and guide your task selection.
 Use this decision tree to route tasks to the correct skill or agent:
 
 ```text
+IF P0 critical bug
+  → /exodia:turbo-fix (13 agents, maximum parallelism)
+
+IF fixing audit findings
+  → /exodia:fix-pipeline (7 agents)
+
 IF struggling > 2 min
   → read metacognitive-guard skill, escalate to deep-think-partner agent
 
@@ -213,7 +179,9 @@ IF multi-agent orchestration needed
     /exodia:eight-gates       - maximum discipline (8 progressive gates, composes all others)
     /exodia:hades             - audited cleanup (3 phases x 4+3 teammates with goggles)
   → exodia commands:
+    /exodia:turbo-fix         - P0 critical bugs (13 agents, maximum parallelism)
     /exodia:fix               - P1/P2/P3 bugs (8 std, 16 max agents)
+    /exodia:fix-pipeline      - fixing findings from an audit (7 agents)
     /exodia:mega-swarm        - codebase audit (6/8/12 agents by mode)
     /exodia:deep-think        - multi-perspective analysis (5 agents)
     /exodia:tournament        - competitive solutions (N+2 agents)
@@ -549,21 +517,11 @@ When you add/rename/remove a plugin:
 
 ### Plugin Structure
 
-Each plugin in `plugins/<name>/` (per [official docs](https://code.claude.com/docs/en/plugins)):
+Each plugin in `plugins/<name>/` (per [official docs](https://code.claude.com/docs/en/plugins)).
 
-```text
-plugins/<name>/
-├── .claude-plugin/
-│   └── plugin.json      # REQUIRED: name, version, description, author
-├── README.md            # REQUIRED: User-facing docs
-├── skills/
-│   └── <skill>/
-│       └── SKILL.md     # YAML frontmatter required
-├── commands/            # Slash commands (.md)
-├── agents/              # Custom agents
-├── hooks/               # Event hooks (hooks.json)
-└── scripts/             # Shell helpers (.sh)
-```
+See `docs/ARCHITECTURE.md` Section 3 for the full plugin structure and passive context layer documentation.
+
+**Required:** `.claude-plugin/plugin.json` (fields: `name`, `version`, `description`, `author`) + `README.md`
 
 ### SKILL.md Format
 
@@ -627,17 +585,10 @@ You MUST NOT:
 
 ## 8. CI & Validation
 
-### CI Jobs (`.github/workflows/ci.yml`)
-
-- `plugin-validation`: `claude plugin validate .` + per-plugin
-- `shell-scripts`: `shellcheck` on `*.sh` files
-- `markdown`: `markdownlint` on `**/*.md`
-- `workflow-syntax`: `actionlint` on workflows
+See `docs/WORKFLOWS.md` for complete CI pipeline documentation, exact shell commands, and the cross-repo `trigger-docs.yml` workflow.
 
 ### Local Validation
 
-Single entrypoint:
-
 ```bash
 ./tooling/scripts/weave-validate.sh
 ```
@@ -743,6 +694,10 @@ claude plugin validate .
 - Specs: `docs/specs/spec-XXXX-*.md`
 - ADRs: `docs/decisions/ADR-XXXX-*.md`
 - Architecture: `docs/ARCHITECTURE.md`
+- Full command reference: `docs/QUICK-REFERENCE.md`
+- CI pipeline details: `docs/WORKFLOWS.md`
+- Plugin creation guide: `docs/PLUGINS.md`
+- Deep engineering principles: `docs/ENGINEERING-PRINCIPLES.md`
 
 ---
 
@@ -778,7 +733,8 @@ You have SUCCEEDED when:
 The following rules are maintained as separate files in `.claude/rules/` and auto-loaded at session start:
 
 - `.claude/rules/solid-principles.md` — SOLID principles for plugin design
-- `.claude/rules/thought-transparency.md` — Observable decision making, task decomposition
+- `.claude/rules/engineering-principles.md` — IF/THEN routing index for Alexander's 26 engineering principles
+- `.claude/rules/thought-transparency.md` — Observable decision making via Processing Log template
 - `.claude/rules/devops-calms.md` — CALMS framework, DORA metrics
 - `.claude/rules/error-handling.md` — Standardized error responses, validation failures