conducty-review

Conducty Review — End-of-Plan Audit

Systematically review all executed prompts from the active plan, verify results with evidence, extract failure patterns for the learning system, and prepare intelligent carry-forward for the next plan.

Workflow

Step 1: Load the Plan

Read the active plan note from the vault (Plans/Plan YYYY-MM-DD HHmm [Topic].md — see [[conducty-obsidian]]). If the user is reviewing a different plan, ask which one and resolve the wikilink.

Parse the prompt queue to identify all prompts and their current status.

Step 2: Identify What Needs Review

• Completed (from checkpoint): Already verified — confirm the checkpoint evidence is in the notes
• Needs-fix with pending fixes: Check if the fix was applied
• Unchecked: Not executed — skip, note as carry-forward
• Already reviewed: Skip unless user requests re-review

Focus on prompts that completed since the last checkpoint or were fixed after checkpoint feedback.

Step 3: Review Each Prompt

For each prompt needing review:

1. Check checkpoint evidence — if the prompt passed checkpoint, confirm the evidence is recorded
2. If no checkpoint evidence: Run [[conducty-verify]] at the prompt's review level
3. Check for unintended changes — git diff/git status (Bash) in the prompt's directory. Any files changed that weren't in the expected outcome?
4. Confirm intent was met — does the actual change match the goal's design, not just the prompt's letter?

Step 4: Record Verdicts

For each prompt, record a verdict:

• completed — changes verified, intent met, no issues
• needs-fix — issues found, can be retried (increment Retries)
• partial — some parts done, rest needs follow-up
• failed — didn't produce useful results, needs rethinking
• blocked — exceeded retries or needs manual intervention

Retry handling for needs-fix:

1. Check Retries count
2. If Retries < 3: invoke [[conducty-debug]] to determine leverage point, generate fix prompt
3. If Retries >= 3: set to blocked with full history of attempts