mercury-audit

Mercury Audit — Entry Point

Step 1 — Confirm the target

Resolve the project directory from the argument or ask. Expand ~. Verify it exists.

Check INCEPTION_API_KEY is set:

echo "${INCEPTION_API_KEY:+set}"

If blank, stop and tell the user to export it.

Step 2 — Ask: Standard or Specialized?

Ask exactly this:

Do you have a specific goal or problem in mind for this audit? — If YES: describe what you're trying to achieve or what isn't working (e.g. "sim trades started at $50, now at $50.28 after 2 days — not locking profits"). — If NO: I'll run a full standard audit.

If the user provides a goal prompt:

• This is a specialized audit. The goal becomes the contract: Mercury will prioritize findings that explain why the system fails to meet that goal, elevating their severity and flagging them as blocks_goal: true.
• Pass the goal verbatim to --goal.

If no goal:

• Standard audit — covers all 10 categories equally.

Step 3 — Create audit branch

BRANCH="audit/mercury-$(date +%Y%m%d-%H%M%S)"
cd "<project_dir>"
git checkout -b "$BRANCH"
echo "branch: $BRANCH"

If the directory is not a git repo, skip branching and note it.

Step 4 — Run the audit

Compute the plugin directory and run the audit script:

SKILL_DIR="$(dirname "${BASH_SOURCE[0]}")"
PLUGIN_DIR="$(cd "$SKILL_DIR/../../.." && pwd)"
AUDIT_SCRIPT="$PLUGIN_DIR/scripts/mercury_audit.py"

Standard:

python3 "$AUDIT_SCRIPT" "<project_dir>" --split-by dir --label "<label>"

Specialized (with goal):

python3 "$AUDIT_SCRIPT" "<project_dir>" --split-by dir --label "<label>" \
  --goal "<user goal verbatim>"

Stream stdout. Reports land in <project_dir>/mercury_<label>_*_report.md. The manifest is at <project_dir>/mercury_<label>_manifest.json.

Step 5 — Read and present findings

Read all mercury_*_report.md files from the manifest. Present inline:

## Mercury Audit — <project> | <branch>
Audit type: Standard | Specialized (goal: "<goal>")

### 🔴 Critical & High Findings
| # | Sev | Blocks Goal | Category | File | Symbol | Issue |
|---|-----|-------------|----------|------|--------|-------|

### Suggested Next Steps & Milestones
<from the report's milestone section>

### Feasibility Verdict
<paragraph>

### Full reports
- <paths>

For specialized audits, surface blocks_goal: true findings at the top, labeled 🎯.

Step 6 — Offer the fix loop

After presenting findings, ask:

Run the fix-and-judge loop on Critical/High findings? I'll spawn a Fixer agent to implement fixes, then judge each one before marking it complete. No fix is closed until I sign off with zero regressions. [Y / N / just the goal-blocking findings]

If the user says yes (or "goal-blocking only"), hand off to the mercury-fix-loop skill with:

• The list of findings to fix (filtered as appropriate)
• The branch name
• The project dir
• The manifest path