@thierrynakoa/fire-flow 12.2.1 → 13.0.1

package/commands/fire-5-handoff.md

@@ -66,6 +66,35 @@ find .planning/phases/ -name "*-RECORD.md" -newer .planning/CONSCIENCE.md
 - Key decisions
 - Issues encountered
 
+### Step 2.5: Context Pruning Discipline (v12.3 — W2-E)
+
+> finding in the wizard creation research: planning context actively **degrades**
+> execution quality. Planning dialogue contains exploratory reasoning, rejected options,
+> and "what if" thinking that pollutes execution focus. Production tools deliberately
+> clear this context before execution.
+
+**Apply the DISTILL-NOT-DUMP principle when writing this handoff:**
+
+```
+WHAT TO CAPTURE (decisions):
+  ✓ Final decisions made — the WHAT that was chosen
+  ✓ Rejected alternatives — WHY the unchosen path was dismissed
+  ✓ Blockers and how they were resolved
+  ✓ State of each file (what it does now)
+  ✓ Next concrete actions
+
+WHAT TO PRUNE (exploration):
+  ✗ Planning dialogue reasoning ("I thought maybe we could...")
+  ✗ Intermediate states that were overwritten
+  ✗ Exploratory code that was discarded
+  ✗ Redundant re-statements of the same decision
+  ✗ Meta-commentary on the development process
+```
+
+**The test:** Could a fresh agent read only this handoff (not the conversation) and
+execute the next session without asking any clarifying questions about already-made
+decisions? If yes — the handoff is correctly pruned. If no — it's missing a decision.
+
 ### Step 3: Compile WARRIOR 7-Step Handoff
 
 Create the handoff document following the W-A-R-R-I-O-R framework:
@@ -176,7 +205,7 @@ next_action: "{what to do next}"
 ### Important Paths:
 - **Project Root:** {path}
 - **Planning Docs:** {path}/.planning/
-- **Skills Library:** ~/.claude/plugins/fire-flow/skills-library/
+- **Skills Library:** ~/.claude/plugins/dominion-flow/skills-library/
 
 ---
 
@@ -273,9 +302,6 @@ next_action: "{what to do next}"
 ---
 
 ## F - Feedback Pairs
-
-> **Research basis (v6.0):** CoALA (Berkeley 2024) specifies episodic memory should store both positive and negative outcomes with structured labels, enabling future agents to avoid known pitfalls and replicate successes.
-
 ### What Worked
 | Pattern | Context | Why It Worked |
 |---------|---------|---------------|
@@ -293,7 +319,6 @@ next_action: "{what to do next}"
 
 ## G — Replay Sequences (v7.0)
 
-> **Research basis:** AgentRR (May 2025) — multi-level record/replay enables
 > both exact replay for identical issues and adapted replay for similar ones.
 
 ### Replayable Resolutions
@@ -308,6 +333,22 @@ next_action: "{what to do next}"
 
 ---
 
+## Session Cost
+
+**Model:** {model used this session — e.g., claude-sonnet-4-6}
+**Estimated spend:** ~${calculated_cost} ({input_tokens} in + {output_tokens} out)
+**Session total:** Rough estimate based on visible tool calls and responses.
+**Cumulative (this milestone):** ~${sum_if_prior_cost_in_handoff} (all sessions combined toward this milestone)
+
+> Pricing reference: Claude Sonnet 4.6 — $3.00/1M input tokens, $15.00/1M output tokens.
+> Estimation method: Count visible tool calls × ~2K tokens each as rough input baseline.
+> Add estimated response length (tokens) as output. Multiply by per-token rate.
+> If a prior Session Cost block exists in the previous handoff, add this session's cost
+> to that cumulative total and record the new running sum.
+> This is an approximation — use Mission Control's AgentBudget panel for precision.
+
+---
+
 ## Quick Resume Commands
 
 ```bash

package/commands/fire-6-resume.md

@@ -96,7 +96,6 @@ next_action: {action}
 
 ### Step 2.5: SDFT On-Policy Self-Distillation (v5.0)
 
-> **Research basis:** Self-Distillation Fine-Tuning (MIT, Shenfeld et al., 2026)
 > — on-policy learning outperforms off-policy by 7 points at 14B params.
 
 **Before reading the full W (Work) section, PREDICT first:**
@@ -177,9 +176,6 @@ Use AskUserQuestion:
 **Skip condition:** If `--quick` flag is set, skip security sweep (user accepts risk).
 
 ### Step 2.8: Load Behavioral Directives (v6.0)
-
-> **Research basis:** CoALA procedural memory + SICA self-modifying agent (ICLR 2025). Active rules guide behavior; proposed rules await confirmation.
-
 ```
 1. Read references/behavioral-directives.md
 2. Extract Active Rules (confidence 3+/5)
@@ -205,34 +201,9 @@ Use AskUserQuestion:
 
 > **Philosophy:** A fresh Claude instance with clean context is the best solver for problems that burned a previous instance. Check for dead ends FIRST — you might solve in 30 seconds what the last instance couldn't in 15 minutes.
 
-**Read `.planning/breadcrumbs/FAILURES.md`** — if it exists, look for entries tagged `[DEAD-END]`:
-
-```
-┌─────────────────────────────────────────────────────────────────────────────┐
-│ DEAD ENDS — {N} unsolved problem(s) from previous sessions                  │
-├─────────────────────────────────────────────────────────────────────────────┤
-│                                                                             │
-│  1. [{priority}] {title}                                                    │
-│     Shelved: {date} by {agent} — Attempts: {N}                             │
-│     Untested hypotheses: {N}                                                │
-│                                                                             │
-└─────────────────────────────────────────────────────────────────────────────┘
-```
-
-**For the highest-priority `[DEAD-END]` entry:**
-
-1. Read the full entry (problem, prior attempts, untested hypotheses, relevant files)
-2. With your fresh context, **attempt the top untested hypothesis**
-3. **If solved:**
-   - Move the entry to `breadcrumbs/LESSONS.md` with `[DEAD-END-SOLVED]` tag
-   - Remove the `[DEAD-END]` tag from `FAILURES.md`
-   - Log the solution approach for future reference
-4. **If still stuck after one focused attempt:**
-   - Update the entry with your new attempt
-   - Don't burn more tokens — move on to normal work
-   - The NEXT fresh instance will try again
+Check the handoff and project state for any previously unsolved problems or blockers from prior sessions. A fresh Claude instance with clean context is often the best solver for problems that burned a previous instance.
 
-**Time budget:** Max 5 minutes on dead-end problems. This is a bonus attempt, not the main work.
+**Time budget:** Max 5 minutes on prior blockers. This is a bonus attempt, not the main work.
 
 **Skip condition:** If `--quick` flag or no `FAILURES.md` exists, skip.
 

package/commands/fire-add-new-skill.md

@@ -48,6 +48,23 @@ Capture and preserve proven solutions when you:
 /fire-add-new-skill --name "retry-backoff" --category "api-patterns"
 ```
 
+## Wizard Step Sequence
+
+```mermaid
+graph LR
+    S1[1: Analyze Source] --> S2[2: Problem + Category]
+    S2 --> S25[2.5: Scope]
+    S25 --> S3[3: Skill Name]
+    S3 --> S4[4: Solution Pattern]
+    S4 --> S5[5: Code Examples]
+    S5 --> S55[5.5: Review & Backtrack]
+    S55 --> S45[4.5: Security Scan]
+    S45 --> S46[4.6: Credential Filter]
+    S46 --> S6[6: Save + Confirm]
+```
+
+> improves LLM step adherence vs. prose. Zero extra tokens.
+
 ## Process
 
 <step number="1">
@@ -99,24 +116,73 @@ STEP 2: CATEGORY SELECTION
 What category best fits this skill?
 
  1. database-solutions   - Database patterns, queries, optimization
+                           Examples: connection pooling, query optimization, migrations
+                           Use if: your skill involves SQL, schemas, or data access
+
  2. api-patterns         - REST, GraphQL, versioning, errors
+                           Examples: rate limiting, retry backoff, error normalization
+                           Use if: your skill is about API design or HTTP integration
+
  3. security             - Auth, validation, encryption
+                           Examples: JWT rotation, input sanitization, RBAC
+                           Use if: your skill prevents a security vulnerability
+
  4. performance          - Caching, optimization, bundles
+                           Examples: lazy loading, memoization, bundle splitting
+                           Use if: your skill measurably reduces load time or memory
+
  5. frontend             - React, Vue, state, CSS
+                           Examples: custom hooks, animation patterns, state machines
+                           Use if: your skill is about UI component logic or styling
+
  6. testing              - Unit, integration, E2E, mocking
+                           Examples: factory fixtures, async test patterns, snapshot traps
+                           Use if: your skill makes tests more reliable or easier to write
+
  7. infrastructure       - Docker, CI/CD, deployment
+                           Examples: multi-stage builds, env injection, health checks
+                           Use if: your skill is about deployment, containerization, or pipelines
+
  8. form-solutions       - Validation, multi-step, uploads
+                           Examples: Zod schemas, file upload patterns, conditional fields
+                           Use if: your skill is about form handling or user input flows
+
  9. ecommerce            - Payments, cart, inventory
+                           Examples: Stripe webhooks, idempotency keys, cart state
+                           Use if: your skill is about commerce transactions or catalog
+
 10. video-media          - Streaming, processing
+                           Examples: HLS chunking, thumbnail generation, CDN patterns
+                           Use if: your skill deals with video/audio/image pipelines
+
 11. document-processing  - PDF, parsing, generation
+                           Examples: PDF generation, markdown parsing, spreadsheet export
+                           Use if: your skill creates or processes documents
+
 12. integrations         - Third-party APIs, webhooks
+                           Examples: OAuth flows, webhook verification, retry on 429
+                           Use if: your skill connects to an external service
+
 13. automation           - Scripts, tasks, workflows
+                           Examples: cron jobs, file watchers, batch processors
+                           Use if: your skill automates a recurring operation
+
 14. patterns-standards   - Design patterns, standards
+                           Examples: factory pattern, observer pattern, naming conventions
+                           Use if: your skill is an architectural or structural pattern
+
 15. methodology          - Process, planning, review
+                           Examples: PR review checklists, estimation strategies, debugging flows
+                           Use if: your skill is about HOW to work, not what to build
+
 16. [custom]             - Create new category
+                           Use if: none of the above fit
 
 Select category (1-16): > [User selection]
 
+> effort 1, impact 2, ratio 2.0). Inline help eliminates misclassification, which
+> degrades skill discoverability for all future users.
+
 -------------------------------------------------------------
 STEP 2.5: SCOPE CLASSIFICATION (v7.0 — SkillRL)
 -------------------------------------------------------------
@@ -288,6 +354,53 @@ difficulty: {easy|medium|hard}
 ```
 </step>
 
+<step number="5.5">
+### Review & Edit All Answers (v12.3 — W2-C)
+
+Before running security and credential scans, display a full review panel so the
+contributor can fix anything without restarting:
+
+```
+=============================================================
+            REVIEW YOUR ANSWERS BEFORE SAVING
+=============================================================
+
+Step-by-step summary of what you've entered:
+
+  1. Problem Description:
+     "{first 120 chars of problem}..."       [Edit]
+
+  2. Category: {selected-category}           [Edit]
+
+  3. Scope: {General | Project}              [Edit]
+
+  4. Skill Name: {skill-name}                [Edit]
+
+  5. Solution Pattern:
+     "{first 120 chars of solution}..."      [Edit]
+
+  6. Code Examples: {N example blocks}       [Edit]
+
+  7. Tags: {tags} | Difficulty: {level}      [Edit]
+
+-------------------------------------------------------------
+
+  [✓ Looks good — run security scan]
+  [Edit Step N — go back and change answer]
+  [Cancel — discard this skill]
+
+=============================================================
+```
+
+**If user selects "Edit Step N":** Return to that step, collect new answer, and
+re-display the review panel with updated values.
+
+**If user selects "Looks good":** Continue to Step 4.5 (Security Scan).
+
+> and react-use-wizard handleStep pattern. Without backtracking, contributors who
+> mis-enter a category must restart from Step 1. High friction → abandonment.
+</step>
+
 <step number="4.5">
 ### Security Scan Gate (MANDATORY)
 
@@ -407,7 +520,7 @@ Unlike the security scan (Step 4.5) which allows "save anyway" for suspicious co
 ### Save and Update Index
 
 1. Save skill file:
-   - Path: `~/.claude/plugins/fire-flow/skills-library/{category}/{skill-name}.md`
+   - Path: `~/.claude/plugins/dominion-flow/skills-library/{category}/{skill-name}.md`
 
 2. Update SKILLS-INDEX.md:
    - Add entry to category section
@@ -416,7 +529,7 @@ Unlike the security scan (Step 4.5) which allows "save anyway" for suspicious co
 
 3. Git commit (if skills library is versioned):
    ```bash
-   cd ~/.claude/plugins/fire-flow/skills-library
+   cd ~/.claude/plugins/dominion-flow/skills-library
    git add {category}/{skill-name}.md SKILLS-INDEX.md
    git commit -m "feat(skills): add {category}/{skill-name}"
    ```
@@ -432,39 +545,45 @@ Display success message:
             SKILL CONTRIBUTED SUCCESSFULLY
 =============================================================
 
-Created: skills-library/{category}/{skill-name}.md
-Updated: skills-library/SKILLS-INDEX.md
+COMPLETION CHECKLIST:
+  [✓] Problem description captured
+  [✓] Category assigned: {category}
+  [✓] Scope: {General | Project}
+  [✓] Skill name: {skill-name}
+  [✓] Solution pattern documented
+  [✓] Code examples: {N} blocks (before + after)
+  [✓] Tags added: {tags}
+  [✓] Difficulty rated: {level}
+  [✓] Security scan: CLEAN
+  [✓] Credential check: CLEAN
+  [✓] Duplicate check: No conflicts
+  [✓] Saved: skills-library/{category}/{skill-name}.md
+  [✓] Index updated: SKILLS-INDEX.md
 
 -------------------------------------------------------------
-SKILL SUMMARY
+SKILL DETAILS
 -------------------------------------------------------------
 
 Name:       {skill-name}
 Category:   {category}
+Scope:      {General | Project}
 Tags:       {tags}
 Difficulty: {difficulty}
 
-This skill is now available for:
-- Search: /fire-search "{skill-name}"
-- Planning: Add to skills_to_apply in BLUEPRINT.md
-- Agents: Automatic reference during execution
-
 -------------------------------------------------------------
 NEXT STEPS
 -------------------------------------------------------------
 
-1. Consider syncing to global library:
-   /fire-skills-sync --push
-
-2. View your contribution:
-   /fire-search --detail {category}/{skill-name}
-
-3. Continue working on current task
+• Search: /fire-search "{skill-name}"
+• Apply:  Add to skills_to_apply in your BLUEPRINT.md
+• Sync:   /fire-skills-sync --push  (publish to global library)
+• View:   /fire-search --detail {category}/{skill-name}
 
 =============================================================
-
-Thank you for contributing to the skills library!
 ```
+
+> Structured completion checklists confirm gates passed and surface next actions,
+> replacing the "did it actually work?" anxiety after multi-step wizard completion.
 </step>
 
 ## Skill Document Template

package/commands/fire-autonomous.md

@@ -47,6 +47,11 @@ arguments:
     type: boolean
     default: false
     description: "Show what would be done without executing"
+
+  --token-efficient:
+    type: boolean
+    default: false
+    description: "Enable ALAS context slicing for ~60% token reduction. Passes --token-efficient to fire-3-execute. Default OFF for best quality."
 ```
 
 ---
@@ -70,6 +75,10 @@ Validate:
   2. VISION.md has phases in range [from_phase, to_phase]
   3. For each phase in range: MEMORY.md exists (PRD/discussion complete)
   4. Path Verification Gate (MANDATORY — working directory matches project)
+  5. Cross-Phase DAG Validation (v13.0 — from fire-3-execute Step 3.55)
+     → Check for circular dependencies between phases
+     → Validate provides/requires contracts match across phase boundaries
+     → STOP if cycles found or contracts broken
 
 IF any phase lacks MEMORY.md:
   Display:
@@ -108,6 +117,8 @@ This gate is NEVER disabled, not even in autonomous mode.
 |    fire-verifier       — 70-point WARRIOR validation          |
 |    fire-reviewer       — 15-persona code review               |
 |                                                                |
+|  Token mode: {IF --token-efficient: EFFICIENT (ALAS) | FULL CONTEXT (best quality)} |
+|                                                                |
 |  Human touchpoints: NONE until completion or blocker           |
 |                                                                |
 |  To interrupt: Ctrl+C or /fire-loop-stop                      |
@@ -178,9 +189,10 @@ FOR phase_number in range(from_phase, to_phase + 1):
 
     Display: "  Executing phase {phase_number}..."
 
-    Run /fire-3-execute {phase_number} --auto-continue --autonomous
+    Run /fire-3-execute {phase_number} --auto-continue --autonomous {IF --token-efficient: --token-efficient}
     // --auto-continue: no breath-boundary interrupts
     // --autonomous: merge gate auto-routes verdicts (v9.0, enhanced v12.0)
+    // --token-efficient: passed through if set (ALAS context slicing, ~60% token reduction)
 
     // ──────────────────────────────────────────────────────
     // Step 3.3: AUTO-VERIFY (read results)
@@ -419,7 +431,7 @@ forward-looking context that the memory system doesn't capture.
 ### Step 8: Sabbath Rest — Session State
 
 ```
-Update .claude/fire-flow.local.md:
+Update .claude/dominion-flow.local.md:
 
 ---
 last_session: {timestamp}

package/commands/fire-complete-milestone.md

@@ -197,7 +197,7 @@ Use AskUserQuestion:
 Update persistent state:
 
 ```markdown
-## .claude/fire-flow.local.md
+## .claude/dominion-flow.local.md
 
 ### Milestone Completion
 - Version: v{version}