Skip to substep 5 correctly communicates jumping past the rest of the git
discovery logic in substep 4 when git repo is not found. Proceed would
suggest normal sequential flow, but we are skipping the conditional branch.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Changed "Skip to substep 6" (which does not exist) to "Proceed to substep 5".
Step only has 5 substeps. After setting NO_GIT flag, workflow continues to
substep 5 (Cross-Reference Story vs Git), not to a non-existent substep 6.
Fixes h2 finding from adversarial review.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
The checklist validates work done DURING step execution.
The NEXT directive is OUTPUT of completion, not a validation criterion.
It happens AFTER the checklist passes, so it does not belong there.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
'Immediately' is implied by HALT. No timing choice exists.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Logical flow: verify checklist → then declare next step
Not: declare next step → then verify checklist
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
- Remove redundant "Do NOT proceed to the next step" (halt already means this)
- Change "item" to "criterion" (more precise terminology)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Correct terminology:
- "Metrics" implies quantitative measurement
- These are actually pass/fail criteria for step completion
- Section is self-validation checklist, not measurement data
Reframe as checkpoint before proceeding to next step:
- Add "Before proceeding to the next step, verify ALL of the following:"
- Change "If any metric" to "If any item"
- Explicit instruction: "Do NOT proceed to the next step" if checklist fails
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
FAILURE MODES section was just inverted SUCCESS METRICS. Not valuable.
Replaced with single catch-all statement: failure to meet any success metric = failure.
Let actual failure modes emerge from usage patterns, not speculation.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Step-01 focus is: load story + discover git changes. Nothing else.
Project context loading belongs in step-04 (Context-Aware Review) where it
provides audit rules, principles, and requirements for validating AC
implementation against project standards.
(See implementation-notes.md for detail)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Step-01 substeps 5:
- If no git changes detected: halt and ask user "Continue anyway?"
Allows AC audit on story File List even if no code changes in git
- Exclude git-ignored files from discrepancy comparison
Prevents false positives if story modified only ignored files
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Step-01 substep-4:
- Use git diff -M to detect renamed/moved files
- Include deleted, renamed files in git_changed_files
- Adversarial reviewer needs to see deletions (e.g., critical code removed)
- Downstream steps will handle these appropriately (documented in implementation-notes)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Substep 3 (Extract File List):
- Removed repetitive wording
- Reference {story_content} variable instead of generic "story file"
- Add error handling: if Dev Agent Record/File List not found, set story_file_list = NO_FILE_LIST
- Consistent with NO_GIT pattern used elsewhere
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Step 01 audit findings:
- Substep 3 was extracting items not needed by step-01 (ACs, tasks, changelog)
Trimmed to only extract story_file_list (needed for git comparison)
- Success/failure criteria now explicitly guard story_content completeness
since downstream steps depend on the full file content
- Removed "downstream" jargon in favor of "later steps"
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Implementation notes for the workflow should be collected in a dedicated
implementation-notes.md file, not embedded in step files. This keeps each
step focused and defers editorial comments to a separate tracking document.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
- Changed priority 1 from exact to resembles: handles format variations (1 2, 1.2, one-two, one thirty two)
- Explicitly prevents false matches: 1-33 does not match 1-32
- Updated priority 3-4 to use resembles instead of contains: supports typos and TTS errors (paiment, passwd)
- Added examples for number variations and compound spoken formats
- Tested with agent validation: handles typos, format variations, misspellings correctly
- Fixed variable naming convention: backticks for names, curlies only for value substitution
- Rewrote Identify Story section with explicit two-path algorithm (file path vs sprint_status search)
- Added verification step for files not in sprint_status with user confirmation flow
- Clarified matching priority order: exact key > full ID > partial > name > description
- Made loopback instructions consistent and explicit (return to user prompt)
- Improved git_discrepancies description from vague "differences" to concrete "mismatches"
- Tested with 30+ test cases and fresh agent review - algorithm is clear and executable
- Add web_bundle: false to frontmatter (workflow needs file access)
- Change "Load and execute" to "Read and follow" (clearer for LLMs)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Change "Story is ready for next work!" to "Code review complete!"
The original phrasing was misleading - when a code review finishes
with status "done", it means the review itself is complete and the
story is marked done in tracking. However, the user may choose to
do additional reviews or the story may genuinely be finished.
"Code review complete" more accurately describes what actually
happened without implying next steps.
- Fix checklist to only accept 'review' status (not 'ready-for-review')
- Include MEDIUM issues in done/in-progress status determination
- Initialize and track fixed_count/action_count variables for summary
- Add sprint-status.yaml sync when story status changes
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude <noreply@anthropic.com>
Replace 'looks good' with `looks good` to avoid nested single quote
issues when IDEs generate command files from workflow YAML.
Co-authored-by: Brian <bmadcode@gmail.com>
Alex Verkhovsky
Nick Pirocanac
Brian Madison
Nguyen Quang Huy