@thierrynakoa/fire-flow 12.2.1 → 13.0.1

package/hooks/run-hook.sh

@@ -1,8 +1,8 @@
-#!/bin/bash
-# MIT License - Copyright (c) 2026 ThierryN - https://github.com/ThierryN/fire-flow
-#
-# Universal hook launcher — works on macOS, Linux, and Windows (Git Bash / WSL)
-# Resolves the actual session-start.sh relative to this script's location.
-
-HOOK_DIR="$(cd "$(dirname "$0")" && pwd)"
-exec bash "$HOOK_DIR/session-start.sh"
+#!/bin/bash
+# MIT License - Copyright (c) 2026 ThierryN - https://github.com/ThierryN/dominion-flow
+#
+# Universal hook launcher — works on macOS, Linux, and Windows (Git Bash / WSL)
+# Resolves the actual session-start.sh relative to this script's location.
+
+HOOK_DIR="$(cd "$(dirname "$0")" && pwd)"
+exec bash "$HOOK_DIR/session-start.sh"

package/hooks/run-session-end.sh

@@ -1,7 +1,7 @@
-#!/bin/bash
-# MIT License - Copyright (c) 2026 ThierryN - https://github.com/ThierryN/fire-flow
-#
-# Universal hook launcher — works on macOS, Linux, and Windows (Git Bash / WSL)
-
-HOOK_DIR="$(cd "$(dirname "$0")" && pwd)"
-exec bash "$HOOK_DIR/session-end.sh"
+#!/bin/bash
+# MIT License - Copyright (c) 2026 ThierryN - https://github.com/ThierryN/dominion-flow
+#
+# Universal hook launcher — works on macOS, Linux, and Windows (Git Bash / WSL)
+
+HOOK_DIR="$(cd "$(dirname "$0")" && pwd)"
+exec bash "$HOOK_DIR/session-end.sh"

package/hooks/session-end.sh

@@ -1,90 +1,90 @@
-#!/bin/bash
-# MIT License - Copyright (c) 2026 ThierryN - https://github.com/ThierryN/fire-flow
-#
-# Dominion Flow Plugin - Session End Hook
-#
-# WHAT THIS DOES:
-#   When a Claude Code session ends, this script runs automatically.
-#   It tries to consolidate session memory into a vector database (Qdrant)
-#   so future sessions can search past experiences semantically.
-#
-# HOW IT WORKS:
-#   1. Checks if the your-memory-repo project exists on this machine
-#   2. Checks if Qdrant (vector database) is running on port 6335
-#   3. Checks if Ollama (local embedding model) is running on port 11434
-#   4. If all three are ready, runs "npm run consolidate" to index new files
-#
-# THIS IS OPTIONAL:
-#   If Qdrant or Ollama are not running, this script exits silently (no error).
-#   The flat .md handoff files in ~/.claude/warrior-handoffs/ are always intact
-#   and used as the fallback. Memory consolidation is a bonus, not a requirement.
-#
-# NOTE FOR NEW USERS:
-#   The MEMORY_PROJECT path below points to a local project that handles
-#   vector indexing. If you do not have this project set up, this hook
-#   will skip silently — your session will still work normally.
-#   To set up vector memory, see: https://github.com/ThierryN/fire-flow
-
-echo "=============================================="
-echo "  DOMINION FLOW - Session End Consolidation"
-echo "=============================================="
-echo ""
-
-MEMORY_PROJECT="$HOME/repos/your-memory-repo"
-
-# =============================================
-# 1. Check if your-memory-repo project exists
-# =============================================
-if [ ! -d "$MEMORY_PROJECT" ]; then
-    echo "[SKIP] your-memory-repo project not found at $MEMORY_PROJECT"
-    echo "       Memory consolidation skipped."
-    exit 0
-fi
-
-# =============================================
-# 2. Check if Qdrant is running (port 6335)
-# =============================================
-if ! curl -s http://localhost:6335/healthz > /dev/null 2>&1; then
-    echo "[SKIP] Qdrant not running on port 6335."
-    echo "       Memory consolidation skipped (no vector DB)."
-    echo "       Flat .md files are still available."
-    exit 0
-fi
-
-# =============================================
-# 3. Check if Ollama is running
-# =============================================
-if ! curl -s http://localhost:11434/api/tags > /dev/null 2>&1; then
-    echo "[SKIP] Ollama not running."
-    echo "       Memory consolidation skipped (no embeddings)."
-    exit 0
-fi
-
-# =============================================
-# 4. Run light consolidation (index new files only)
-# =============================================
-echo ">>> Running light memory consolidation..."
-echo "    (Index new/changed files only — fast, <30s)"
-echo ""
-
-cd "$MEMORY_PROJECT" || exit 0
-
-# Run the consolidate command (indexes new files + extracts facts)
-npm run consolidate 2>&1 | tail -20
-
-CONSOLIDATE_EXIT=$?
-
-if [ $CONSOLIDATE_EXIT -eq 0 ]; then
-    echo ""
-    echo ">>> Memory consolidation complete."
-else
-    echo ""
-    echo "[WARN] Consolidation exited with code $CONSOLIDATE_EXIT"
-    echo "       This is non-fatal. Memory files are still intact."
-fi
-
-echo ""
-echo "=============================================="
-echo "  SESSION END CONSOLIDATION COMPLETE"
-echo "=============================================="
-echo ""
+#!/bin/bash
+# MIT License - Copyright (c) 2026 ThierryN - https://github.com/ThierryN/dominion-flow
+#
+# Dominion Flow Plugin - Session End Hook
+#
+# WHAT THIS DOES:
+#   When a Claude Code session ends, this script runs automatically.
+#   It tries to consolidate session memory into a vector database (Qdrant)
+#   so future sessions can search past experiences semantically.
+#
+# HOW IT WORKS:
+#   1. Checks if the your-memory-repo project exists on this machine
+#   2. Checks if Qdrant (vector database) is running on port 6335
+#   3. Checks if Ollama (local embedding model) is running on port 11434
+#   4. If all three are ready, runs "npm run consolidate" to index new files
+#
+# THIS IS OPTIONAL:
+#   If Qdrant or Ollama are not running, this script exits silently (no error).
+#   The flat .md handoff files in ~/.claude/warrior-handoffs/ are always intact
+#   and used as the fallback. Memory consolidation is a bonus, not a requirement.
+#
+# NOTE FOR NEW USERS:
+#   The MEMORY_PROJECT path below points to a local project that handles
+#   vector indexing. If you do not have this project set up, this hook
+#   will skip silently — your session will still work normally.
+#   To set up vector memory, see: https://github.com/ThierryN/dominion-flow
+
+echo "=============================================="
+echo "  DOMINION FLOW - Session End Consolidation"
+echo "=============================================="
+echo ""
+
+MEMORY_PROJECT="$HOME/repos/your-memory-repo"
+
+# =============================================
+# 1. Check if your-memory-repo project exists
+# =============================================
+if [ ! -d "$MEMORY_PROJECT" ]; then
+    echo "[SKIP] your-memory-repo project not found at $MEMORY_PROJECT"
+    echo "       Memory consolidation skipped."
+    exit 0
+fi
+
+# =============================================
+# 2. Check if Qdrant is running (port 6335)
+# =============================================
+if ! curl -s http://localhost:6335/healthz > /dev/null 2>&1; then
+    echo "[SKIP] Qdrant not running on port 6335."
+    echo "       Memory consolidation skipped (no vector DB)."
+    echo "       Flat .md files are still available."
+    exit 0
+fi
+
+# =============================================
+# 3. Check if Ollama is running
+# =============================================
+if ! curl -s http://localhost:11434/api/tags > /dev/null 2>&1; then
+    echo "[SKIP] Ollama not running."
+    echo "       Memory consolidation skipped (no embeddings)."
+    exit 0
+fi
+
+# =============================================
+# 4. Run light consolidation (index new files only)
+# =============================================
+echo ">>> Running light memory consolidation..."
+echo "    (Index new/changed files only — fast, <30s)"
+echo ""
+
+cd "$MEMORY_PROJECT" || exit 0
+
+# Run the consolidate command (indexes new files + extracts facts)
+npm run consolidate 2>&1 | tail -20
+
+CONSOLIDATE_EXIT=$?
+
+if [ $CONSOLIDATE_EXIT -eq 0 ]; then
+    echo ""
+    echo ">>> Memory consolidation complete."
+else
+    echo ""
+    echo "[WARN] Consolidation exited with code $CONSOLIDATE_EXIT"
+    echo "       This is non-fatal. Memory files are still intact."
+fi
+
+echo ""
+echo "=============================================="
+echo "  SESSION END CONSOLIDATION COMPLETE"
+echo "=============================================="
+echo ""

package/hooks/session-start.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
-# MIT License - Copyright (c) 2026 ThierryN - https://github.com/ThierryN/fire-flow
+# MIT License - Copyright (c) 2026 ThierryN - https://github.com/ThierryN/dominion-flow
 #
 # Dominion Flow Plugin - Session Start Hook
 #

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@thierrynakoa/fire-flow",
-  "version": "12.2.1",
-  "description": "Dominion Flow — the most comprehensive orchestration platform for Claude Code. 46 commands, 15 agents, 478+ skills. Plan → Execute → Verify → Handoff with parallel execution, session memory, circuit breaker safety, Phoenix Rebuild, and learning mode.",
+  "version": "13.0.1",
+  "description": "Dominion Flow — the most comprehensive orchestration platform for Claude Code. 46 commands, 15 agents, 400+ skills. Plan, Execute, Verify, Handoff with parallel execution, session memory, circuit breaker safety, Phoenix Rebuild, and learning mode.",
   "keywords": [
     "claude-code",
     "dominion-flow",
@@ -19,35 +19,15 @@
   "license": "MIT",
   "repository": {
     "type": "git",
-    "url": "https://github.com/ThierryN/fire-flow.git"
+    "url": "git+https://github.com/ThierryN/fire-flow.git"
   },
   "homepage": "https://github.com/ThierryN/fire-flow#readme",
   "bugs": {
     "url": "https://github.com/ThierryN/fire-flow/issues"
   },
   "bin": {
-    "fire-flow": "./bin/cli.js"
+    "fire-flow": "bin/cli.js"
   },
-  "files": [
-    "bin/",
-    "agents/",
-    "commands/",
-    "hooks/",
-    "skills-library/",
-    "templates/",
-    "references/",
-    "workflows/",
-    "plugin.json",
-    ".claude-plugin/",
-    "README.md",
-    "QUICK-START.md",
-    "COMMAND-REFERENCE.md",
-    "DOMINION-FLOW-OVERVIEW.md",
-    "ARCHITECTURE-DIAGRAM.md",
-    "TROUBLESHOOTING.md",
-    "validation-config.yml",
-    "version.json"
-  ],
   "engines": {
     "node": ">=16.0.0"
   }

package/plugin.json

@@ -1,7 +1,7 @@
-{
-  "name": "fire-flow",
-  "version": "12.2.0",
-  "description": "Dominion Flow is a comprehensive orchestration platform that empowers your Claude agent. It gives Claude a complete, structured workflow for taking projects from idea to production — with parallel execution, session memory across handoffs, 70-point quality verification, autonomous mode, Phoenix Rebuild (messy-to-clean), E2E testing, and a circuit breaker to prevent stuck loops. Includes 46 slash commands organized into 8 tiers covering planning, execution, verification, debugging, skills management, milestone tracking, and learning mode.",
-  "commands": "commands/",
-  "agents": "agents/"
-}
+{
+  "name": "dominion-flow",
+  "version": "12.9.0",
+  "description": "Dominion Flow is a comprehensive orchestration platform that empowers your Claude agent. It gives Claude a complete, structured workflow for taking projects from idea to production — with parallel execution, session memory across handoffs, 70-point quality verification, autonomous mode, Phoenix Rebuild (messy-to-clean), E2E testing, and a circuit breaker to prevent stuck loops. Includes 50 slash commands organized into 8 tiers covering planning, execution, verification, debugging, skills management, milestone tracking, and learning mode.",
+  "commands": "commands/",
+  "agents": "agents/"
+}

package/references/autonomy-levels.md

@@ -0,0 +1,235 @@
+# Configurable Autonomy Levels
+
+> Graduated autonomy from fully supervised to fully autonomous — users choose how much control to retain at each decision point.
+
+> agents need configurable autonomy, not binary on/off. Microsoft AutoGen "Human-in-the-Loop
+> Patterns" (ICML 2024) — graduated autonomy with explicit checkpoints reduces user anxiety
+> and increases adoption by 45%. Manus AI's "confidence-gated escalation" pattern — let the
+> agent self-assess when to ask for help vs. proceed autonomously.
+
+---
+
+## Overview
+
+Previous versions offered two modes: `--autonomous` (no checkpoints) and `--manual` (all checkpoints). This was too coarse — users either got interrupted too often or had zero visibility.
+
+Configurable autonomy introduces **4 levels** that control exactly which decision points require human input.
+
+---
+
+## Four Autonomy Levels
+
+### Level 1: SUPERVISED (equivalent to old `--manual`)
+
+Every decision point pauses for human input.
+
+```
+Human input required at:
+  ✓ Plan approval (BLUEPRINT.md review)
+  ✓ Breath completion (approve before next breath)
+  ✓ Merge gate (review combined verdict)
+  ✓ Skill extraction (approve new skills)
+  ✓ Gap closure (decide: fix vs. accept vs. re-plan)
+  ✓ Phase completion (approve verification report)
+  ✓ Handoff content (review before saving)
+
+Best for:
+  - Learning how Dominion Flow works
+  - Security-sensitive projects
+  - First time using the plugin
+  - Projects where mistakes are expensive
+```
+
+### Level 2: GUIDED (new — recommended default)
+
+Pauses only at major decision points. Auto-routes routine decisions.
+
+```
+Human input required at:
+  ✓ Plan approval (BLUEPRINT.md review)
+  ✗ Breath completion (auto-advance if no errors)
+  ✓ Merge gate ONLY if verdict is REJECTED/BLOCK
+  ✗ Skill extraction (auto-create, auto-scan)
+  ✓ Gap closure (human decides: fix vs. accept)
+  ✓ Phase completion (review verification report)
+  ✗ Handoff content (auto-save)
+
+Auto-routed:
+  - Successful breath → next breath (logged)
+  - APPROVED/CONDITIONAL merge → continue (logged)
+  - Skill extraction → auto-create with security scan
+  - Handoff → auto-save to warrior-handoffs/
+
+Best for:
+  - Day-to-day development
+  - Projects you understand well
+  - Balanced speed and oversight
+```
+
+### Level 3: AUTONOMOUS (equivalent to old `--autonomous`)
+
+Only stops for blockers and failures. Maximum speed.
+
+```
+Human input required at:
+  ✗ Plan approval (auto-approve if plan follows template)
+  ✗ Breath completion (auto-advance always)
+  ✓ Merge gate ONLY if REJECTED (critical failures)
+  ✗ Skill extraction (auto-create, auto-scan)
+  ✗ Gap closure (auto-fix minor, escalate critical)
+  ✗ Phase completion (auto-approve if score ≥ 70%)
+  ✗ Handoff content (auto-save)
+
+Only stops for:
+  - Circuit breaker TRIPPED
+  - Verification score < 50% (critical failure)
+  - BLOCKED state (external dependency)
+  - Security finding (critical severity)
+
+Best for:
+  - Experienced users
+  - Well-defined projects with good test coverage
+  - /fire-autonomous and /fire-loop workflows
+  - Overnight or unattended execution
+```
+
+### Level 4: SENTINEL (new — maximum autonomy with safety)
+
+Full autonomy with a safety net. The agent runs completely unattended but
+maintains a log of all decisions for post-hoc review.
+
+```
+Human input required at:
+  ✗ Everything auto-routed
+  ✓ ONLY stops for circuit breaker TRIPPED
+
+Safety net:
+  - ALL auto-decisions logged to .planning/sentinel-log.md
+  - Confidence < 50% decisions flagged for review (but not paused)
+  - Git tags created at each phase boundary for easy rollback
+  - /fire-cost auto-runs every 3 breaths
+  - Handoff auto-saved every 30 minutes
+
+Post-hoc review:
+  After completion, review .planning/sentinel-log.md:
+  | Time | Decision Point | Auto-Action | Confidence | Flag |
+  | ... | Merge gate | Auto-proceed | 85% | - |
+  | ... | Gap closure | Auto-fixed | 72% | - |
+  | ... | Skill extract | Auto-created | 90% | - |
+  | ... | Phase verify | Auto-approved 68/70 | 65% | ⚠ LOW |
+
+Best for:
+  - Prototyping and experimentation
+  - Tasks where speed matters more than precision
+  - Projects with comprehensive test suites
+  - Users who prefer review-after over approve-before
+```
+
+---
+
+## Configuration
+
+Set autonomy level per-project in `.planning/config.yml`:
+
+```yaml
+autonomy:
+  level: guided  # supervised | guided | autonomous | sentinel
+
+  # Override specific decision points regardless of level:
+  overrides:
+    plan_approval: always_pause      # always_pause | auto_approve | confidence_gated
+    merge_gate: on_failure_only      # always_pause | on_failure_only | auto_route
+    skill_extraction: auto_with_scan # always_pause | auto_with_scan | skip
+    phase_completion: always_pause   # always_pause | score_gated | auto_approve
+```
+
+Set via command flags (per-session override):
+
+```bash
+# Set level for this execution
+/fire-3-execute 2 --autonomy supervised
+/fire-3-execute 2 --autonomy guided
+/fire-3-execute 2 --autonomy autonomous
+/fire-3-execute 2 --autonomy sentinel
+
+# Legacy flags still work (mapped to levels)
+/fire-3-execute 2 --manual        # → supervised
+/fire-3-execute 2 --autonomous    # → autonomous
+```
+
+---
+
+## Confidence-Gated Escalation
+
+At any autonomy level, the agent can escalate to human input when confidence is low:
+
+```
+FOR each auto-routed decision:
+
+  confidence = assess_confidence(decision_context)
+
+  IF confidence >= 80%:
+    → Auto-route. Log decision.
+
+  IF confidence 50-79% AND level >= AUTONOMOUS:
+    → Auto-route. Log with ⚠ flag for post-hoc review.
+
+  IF confidence 50-79% AND level < AUTONOMOUS:
+    → Pause for human input. Show confidence and reasoning.
+
+  IF confidence < 50% (any level):
+    → Pause for human input. Show full analysis.
+    → Exception: SENTINEL level still auto-routes but flags for review.
+```
+
+---
+
+## Decision Point Reference
+
+| Decision Point | SUPERVISED | GUIDED | AUTONOMOUS | SENTINEL |
+|---------------|-----------|--------|-----------|---------|
+| Plan approval | Pause | Pause | Auto | Auto |
+| Breath advance | Pause | Auto | Auto | Auto |
+| Merge gate (pass) | Pause | Auto | Auto | Auto |
+| Merge gate (fail) | Pause | Pause | Pause | Auto + flag |
+| Skill extraction | Pause | Auto + scan | Auto + scan | Auto + scan |
+| Gap closure (minor) | Pause | Pause | Auto-fix | Auto-fix |
+| Gap closure (critical) | Pause | Pause | Pause | Auto + flag |
+| Phase completion | Pause | Pause | Auto (≥70%) | Auto |
+| Handoff save | Pause | Auto | Auto | Auto |
+| Circuit break | Stop | Stop | Stop | Stop |
+| Security critical | Stop | Stop | Stop | Stop + flag |
+
+---
+
+## Integration
+
+### /fire-3-execute
+
+The `--autonomy` flag replaces the binary `--autonomous`/`--manual` flags. Legacy flags are mapped:
+- `--manual` → `--autonomy supervised`
+- `--autonomous` → `--autonomy autonomous`
+- No flag → uses `.planning/config.yml` setting, default: `guided`
+
+### /fire-autonomous
+
+Always runs at `autonomous` level. Can be upgraded to `sentinel` with `--sentinel` flag.
+
+### /fire-loop
+
+Inherits autonomy level from the triggering command. Displays current level in loop status.
+
+### /fire-4-verify
+
+Phase completion approval follows the autonomy level's `phase_completion` setting.
+
+---
+
+## References
+
+- **Inspiration:** Devin's guardrails system, Manus AI confidence-gated escalation
+- **Related:** `references/circuit-breaker.md` — hard stops override all autonomy levels
+- **Related:** `references/error-classification.md` — BLOCKED state overrides all autonomy levels
+- **Related:** `commands/fire-cost.md` — auto-runs in SENTINEL mode
+- **Consumer:** `commands/fire-3-execute.md` — primary integration (Step 8.75)
+- **Consumer:** `commands/fire-autonomous.md` — default AUTONOMOUS level

package/references/behavioral-directives.md

@@ -151,7 +151,6 @@ NEW INSIGHT discovered during task resolution
 
 ### Hard Admissibility Check (v7.0 — MPR)
 
-> **Research basis:** MPR (Sep 2025) — Predicate-form rules with hard admissibility checks
 > prevent known-bad actions before they execute, not after.
 
 ```
@@ -297,7 +296,6 @@ No data yet — v8.0 just deployed. First 5 gate outcomes will establish baselin
 
 ## Formal Constraints (v11.0 — AgentSpec-Style Enforcement)
 
-> **Research basis (v11.0):** AgentSpec DSL (ICSE 2026) — machine-checkable ALWAYS/NEVER/WHEN-THEN
 > constraints that are enforced at runtime, not just documented. Unlike predicate rules (which
 > guide behavior), formal constraints BLOCK execution if violated.
 
@@ -359,7 +357,101 @@ THEN: fix_findings BEFORE presenting_to_human
 
 ---
 
-*Dominion Flow v11.0 — MPR predicate rules + HAC enforcement + review gate + simplicity + version tracking + formal constraints*
+## Frequency-Ranked Core Learnings (v13.0 — Nodeland/Beads Pattern)
+
+> rank behavioral rules by application frequency. Most-used rules are injected first into
+> context (highest attention weight), rarely-used rules are demoted. This is natural selection
+> for rules: useful ones survive, useless ones fade.
+
+### How It Works
+
+```
+EVERY 10 sessions (or when `Last applied` dates are updated):
+
+  1. RANK all Active Rules by recency * frequency:
+     score = (1 / days_since_last_applied) * total_applications
+
+  2. TOP 5 rules → "Core Learnings" (injected into every executor prompt)
+  3. Rules 6-12 → "Active" (available via DTG lookup, not pre-injected)
+  4. Rules with 0 applications in 30 days → "Dormant" (candidate for retirement)
+
+  FORMAT for Core Learnings injection:
+    <core_learnings>
+    1. [R1] Verify working directory before file operations
+    2. [R4] Read files before editing
+    3. [R7] Eliminate hypotheses with evidence before fixing
+    4. [R5] Run tests after every code change
+    5. [R2] Use parameterized queries for user input
+    </core_learnings>
+```
+
+### Current Rankings (v13.0 baseline)
+
+| Rank | Rule | Last Applied | Status |
+|------|------|-------------|--------|
+| 1 | R1 (path verification) | 2026-02-21 | Core |
+| 2 | R4 (read before edit) | 2026-02-21 | Core |
+| 3 | R7 (evidence-based debug) | 2026-02-21 | Core |
+| 4 | R5 (tests after change) | 2026-02-21 | Core |
+| 5 | R3 (check existing code) | 2026-02-21 | Core |
+| 6-12 | R2, R6, R8-R12 | varies | Active |
+
+### Dormancy Protocol
+
+```
+IF rule.last_applied > 30 days ago AND rule.total_applications < 3:
+  → Move to "Dormant" section
+  → Log: "Rule {N} dormant — no applications in 30 days"
+  → Dormant rules are NOT deleted — they can be reactivated if relevant again
+  → Retirement requires explicit review (dormancy ≠ retirement)
+```
+
+---
+
+## CNCF Four Pillars Taxonomy (v13.0 — Rule Classification)
+
+> into four pillars: Observability, Governance, Security, and Lifecycle. Mapping our rules
+> to these pillars reveals coverage gaps and makes rule selection faster at decision time.
+
+### Pillar Mapping
+
+| Pillar | Rules | Anti-Patterns | Formal Constraints |
+|--------|-------|---------------|-------------------|
+| **Observability** | R5 (tests after change), R10 (compaction signals), R11 (review gate) | AP2 (silent errors), AP3 (retry loops) | ALWAYS: tests_pass, WHEN: output_decline |
+| **Governance** | R6 (granular commits), R12 (no single-use abstractions) | AP5 (over-engineering) | ALWAYS: git_status_clean, WHEN: reviewer_verdict |
+| **Security** | R1 (path verification), R2 (parameterized queries), R4 (read before edit) | AP1 (modify tests to pass) | NEVER: commit_credentials, NEVER: delete_without_reading, ALWAYS: working_directory |
+| **Lifecycle** | R3 (check existing code), R7 (evidence-based debugging), R8 (check deps), R9 (search skills) | AP4 (Qdrant error shape) | WHEN: confidence < 30, WHEN: circuit_breaker |
+
+### Coverage Analysis
+
+```
+Observability:  5 rules/constraints — GOOD coverage
+Governance:     4 rules/constraints — ADEQUATE
+Security:       6 rules/constraints — STRONG coverage
+Lifecycle:      6 rules/constraints — GOOD coverage
+
+Gap: No OBSERVABILITY rule for monitoring context usage
+  → Candidate: "IF context exceeds 70% THEN log warning + trigger tier assessment"
+  → Status: PROPOSED (needs 3 confirmations to activate)
+```
+
+### Decision-Time Lookup
+
+When DTG (Decision-Time Guidance) fires at a decision point, use pillar classification for fast rule lookup:
+
+```
+TOOL_CHOICE decision     → check: Lifecycle pillar rules
+ARCHITECTURE decision    → check: Governance + Lifecycle
+ERROR_HANDLING decision  → check: Observability + Security
+SECURITY decision        → check: Security pillar rules (ALL)
+PERFORMANCE decision     → check: Governance (no over-engineering)
+SCOPE decision           → check: Governance + Lifecycle
+```
+
+---
+
+*Dominion Flow v13.0 — MPR predicate rules + HAC enforcement + review gate + simplicity + version tracking + formal constraints + CNCF taxonomy*
 *v7.0: Predicate format + HAC enforcement (2026-02-21)*
 *v8.0: Rules 11-12, Anti-Pattern 5, Version Performance Registry (2026-02-21)*
 *v11.0: AgentSpec formal ALWAYS/NEVER/WHEN-THEN constraints (2026-03-01)*
+*v13.0: CNCF Four Pillars taxonomy mapping (2026-03-12)*

package/references/blocker-tracking.md

@@ -14,7 +14,7 @@ Centralized blocker tracking across all phases and plans. Blockers are obstacles
 .planning/BLOCKERS.md
 ```
 
-Created automatically by `/fire-1-new`. Updated by execution commands when blockers are encountered.
+Created automatically by `/fire-1a-new`. Updated by execution commands when blockers are encountered.
 
 ---