Provides a structured workflow for analyzing CRAFT framework issues, improvements, and questions
You raise an issue; the AI reframes it for optimal analysis, provides confidence-scored findings grounded in project evidence, double-checks for hallucination, searches closed subproject precedent, and routes approved actions to immediate execution, the project roadmap, or a new subproject.
CRAFT Issue Analyzer & Roadmap Router
Tags: Introduced in Beta, Issue, Analysis, Roadmap, Confidence, Hallucination, QA, Subproject, Cowork
TL;DR
What It Does
Provides a structured workflow for analyzing CRAFT framework issues, improvements, and questions. You raise an issue; the AI reframes it for optimal analysis, provides confidence-scored findings grounded in project evidence, double-checks for hallucination, searches closed subproject precedent, and routes approved actions to immediate execution, the project roadmap, or a new subproject.
How It Works
The AI captures your issue, classifies it (bug, improvement, question, incompatibility, or feature request), and transforms it into an analysis frame with a precise root question, scope boundaries, evidence requirements, and success criteria. Analysis uses a four-tier evidence hierarchy — CRAFT project files (highest), observed behavior, design intent, and AI reasoning (lowest). Every finding gets a confidence score (0–100) and a hallucination check. Closed subprojects are searched for historical precedent. Results are presented with recommendations, then routed based on your decision.
What To Expect
A structured analysis report with confidence-scored findings in 10–30 minutes depending on issue complexity and evidence availability. Clear recommendations with effort estimates, plus automatic roadmap and git checkpoint updates after you approve.
Best Results When You
Describe your issue clearly — the more specific, the better the analysis frame. Issues with concrete examples (“CWK-001 skips Step 5 when files are present”) produce higher-confidence findings than vague observations. Review the reframed analysis before the AI proceeds.
Typical Time
10–30 minutes. Simple questions resolve in 10 minutes; complex bugs with multiple evidence sources may take 20–30 minutes.
Difficulty
Intermediate — you describe the issue and review the analysis. The AI handles reframing, evidence gathering, hallucination checking, and roadmap routing. You make the final routing decision.
How To Start
Describe the issue, question, or improvement you want analyzed. The AI handles the structured analysis pipeline.
STEP 1Describe the Issue
Tell the AI what you have observed, what question you have, or what improvement you want to explore. The AI classifies the issue type and reframes it for structured analysis.
Available parameters
-
issue_type
· string · optional · default: “auto-detect”
Type of issue. Options: “bug,” “improvement,” “question,” “incompatibility,” “feature_request.” Auto-detected from your description if not specified. -
project_id
· string · required
Three-digit project ID for roadmap routing. -
roadmap_file
· string · optional · default: “auto-detect”
Path to project roadmap .docx file. Auto-detected from workspace if not specified.
Invocations
Standard Analysis
#H->AI::Directive: (Analyze CRAFT issue and route to roadmap with project_id=067)
Typed Analysis
#H->AI::Directive: (Analyze CRAFT issue and route to roadmap with project_id=067, issue_type=incompatibility)
STEP 2Review the Reframed Analysis
The AI presents an analysis frame with root question, scope boundaries, evidence needed, and success criteria. Confirm or adjust before the AI proceeds with evidence gathering.
STEP 3Review Findings and Route
The AI presents confidence-scored findings, hallucination check results, and closed-subproject precedent. If overall confidence is 60 or above, recommendations are offered. You choose: execute now, defer to roadmap, split (some now, some later), or open a new subproject for multi-session scope.
Usage Examples
Two scenarios — a bug investigation with immediate fix, and an improvement idea routed to a new subproject.
Example 1 · Bug investigation with immediate fix
User:
CWK-001 seems to skip the COM activation status message when the project file has more than 10 active priorities.
AI:
Classified as: bug. Reframing for analysis. Root question: “What causes CWK-001 to skip COM activation status when the project file has more than 10 active priorities?” Does this capture your concern?
User:
Yes, proceed.
AI:
Analysis complete. 3 findings at confidence 78/100. No hallucination flags. No closed-subproject precedent. Recommendation R-1: Add priority count guard in Step 7 COM block (effort: low). Execute now, defer, or split?
User:
Now.
AI:
R-1 executed. Git checkpoint saved. Roadmap updated with completed item.
Example 2 · Improvement routed to new subproject
User:
We need a way to batch-validate all recipes after a major cookbook refactor.
AI:
Classified as: feature_request. 2 closed-subproject precedent hits (advisory). 4 recommendations, total effort: high, estimated 3+ sessions. Suggesting SP_NEW routing — scope warrants its own subproject.
User:
SP_NEW.
AI:
Routing to CWK-048 with analysis findings as seed material. Roadmap cross-reference entry added.
How AI Reads This Recipe
When this recipe is triggered, the AI runs a structured 8-step issue analysis pipeline with evidence-based reasoning and hallucination safeguards. The AI should:
- CAPTURE the issue verbatim, classify its type (auto-detect or use provided type), and report the classification.
- REFRAME the issue into an analysis frame with root question, scope boundaries, evidence requirements, and success criteria. Present to user for confirmation before proceeding.
- GATHER evidence using the four-tier hierarchy: Tier 1 (CRAFT project files, highest confidence), Tier 2 (observed behavior), Tier 3 (design intent, including closed-subproject precedent), Tier 4 (AI reasoning, lowest confidence). Score each finding 0–100.
- RUN the hallucination check on every finding: verify file references exist, confirm claims were actually read (not reconstructed), check for Lessons Learned conflicts, and flag untested assumptions. Recalculate confidence after checks.
- SEARCH closed subprojects for precedent: glob project/subprojects/_closed/*/STATE.txt, keyword-match against the analysis frame, surface top 3 hits as advisory Tier-3 evidence capped at confidence 65.
- PRESENT findings with confidence scores. Offer recommendations only if overall confidence is 60 or above.
- ROUTE based on user decision: NOW (execute immediately), LATER (defer to roadmap), SPLIT (partial execution), SP_NEW (new subproject via CWK-048, blocked if an active subproject already exists per single-active rule).
- UPDATE the roadmap file and run a git checkpoint after any file modifications.
The hallucination check is not optional — it must run on every finding regardless of confidence level. The closed-subproject search should fail gracefully if _closed/ does not exist. SP_NEW routing must respect the single-active subproject rule.
When to Use This Recipe
Use this recipe when you:
- Discover a bug, inconsistency, or unexpected behavior in CRAFT.
- Have an improvement idea that needs structured analysis before acting on it.
- Want confidence-scored answers to CRAFT questions grounded in project evidence, not guesswork.
- Need to route an issue to the project roadmap with proper tracking and traceability.
- Have an issue large enough to warrant its own subproject.
Do not use this recipe when:
You already know exactly what to change and it takes less than 5 minutes. Running the full 8-step analysis pipeline on a one-line typo fix or a known-answer question wastes time. Just make the fix and run CWK-005 for a git checkpoint. Use this recipe when the issue needs investigation, evidence gathering, or involves uncertainty about the right approach.
Recipe FAQ
Q.What are the confidence scoring rules?
90–100: directly confirmed by CRAFT spec or observed behavior. 70–89: supported by multiple files plus consistent logic. 50–69: one source plus reasonable inference. 30–49: pattern matching or analogical reasoning. 11–29: speculative, flagged as needs-verification. 0–10: insufficient evidence.
Q.What happens if the hallucination check drops a finding below 10?
The finding is marked UNVERIFIED and excluded from recommendations. The AI suggests external research or testing to confirm before acting on it.
Q.What is the closed-subproject precedent search?
The AI searches project/subprojects/_closed/ for historically closed subprojects whose title, goal, or completion criteria match your current issue. Hits surface as advisory evidence (not binding) capped at confidence 65, helping you learn from past decisions and pitfalls.
Q.When should I use SP_NEW instead of NOW or LATER?
SP_NEW is appropriate when recommendations total effort is high, span 3 or more sessions, touch 4 or more distinct files or recipes, or map to a multi-phase initiative. The AI may proactively suggest SP_NEW when these heuristics are met.
Q.What if I already have an active subproject?
SP_NEW is blocked by the single-active subproject rule. You can close the existing subproject first via CWK-049, defer the issue to roadmap (LATER), or execute partial changes now (SPLIT).
Q.Does this recipe modify files directly?
Only when you choose NOW or SPLIT. Executed changes are followed by a git checkpoint. LATER and SP_NEW only update the roadmap without modifying source files.
Version History
Changes to this recipe over time. Most recent first.
v1.01a
2026-03-15
Added SP_NEW as 5th action routing option — issues large enough for their own subproject route to CWK-048 with analysis findings as seed material. Added closed-subproject precedent search: globs _closed/*/STATE.txt, keyword-matches against analysis frame, surfaces top 3 hits as advisory Tier-3 evidence. SP_NEW respects single-active rule. Roadmap cross-reference entry for subproject-routed items.
v1.00a
2026-02-28
Initial release. 8-step analysis pipeline: issue intake, reframe, evidence-based analysis with 4-tier hierarchy, hallucination check, results presentation, human confirmation, action routing (NOW / LATER / SPLIT), roadmap update with git checkpoint.
Get this recipe with CRAFT for Claude Cowork
Cowork recipes ship bundled with CRAFT for Claude Cowork — there’s no separate download. Clone the framework once, and your AI runs every recipe automatically when invoked.
Pull anytime to stay on the latest version — free to clone, no login or email required.
Then start your session
Once CRAFT is in your project folder, open a new Cowork session and ask Claude to initialize. For example:
You
Please initialize my CRAFT session.
Claude
CRAFT session ready. Your project is loaded, your persona is active, and your recipes are available. What would you like to work on?
Not familiar with Git? Download as a ZIP
No command line needed. Just download, move, and unzip:
- Open the CRAFT framework repo on GitHub.
- Click the green Code button, then choose Download ZIP.
- Move the downloaded ZIP into your Claude Cowork project folder.
- Unzip it: double-click on Mac, or right-click → Extract All on Windows.
