history

name	history
description	List past Crucible reviews stored in .review/reports/ and optionally show the full markdown of one. Use to see how the project's review verdicts have evolved over time.

/crucible:history — Review History

List past Crucible reviews and offer to show one in full. Reviews are stored at .review/reports/<review_id>.md in the project root.

Steps

1. List reports. Find every file under .review/reports/ matching the pattern *.md. If none exist, print:

   No past reviews found. Run /crucible to create one.
   

   and exit.

2. Parse metadata from each report. For each file, extract:

   • review_id — from the filename (strip .md).
   • final_score — from the line beginning with **Score:** (or fall back to Score: if formatting differs).
   • final_verdict — from the line beginning with **Verdict:**.
   • scope_description — from the H1 line # Crucible Review — <description>.
   • completed_at — from the metadata line _Review ID: ... · Generated: <timestamp>_.

   If parsing fails on any field, fall back to ? for that field.

3. Sort reports newest-first by completed_at if available, else by filename (which encodes the timestamp prefix).

4. Print a table. Use fixed-width columns. Truncate long descriptions to 40 chars with ….

   #   Review ID                          Score  Verdict                Scope
   ─────────────────────────────────────────────────────────────────────────────────
   1   2026-05-10-1430-auth-refactor      7.1    conditional_approval   auth module rewrite
   2   2026-05-09-1620-api-rewrite        8.4    approved               /api/v2 endpoints
   ...
   

5. Prompt for selection.

   Show full report for which # (or `n` to exit)?
   

   • If the user enters a number that maps to a row, print the full contents of that report file.
   • If the user enters n, no, exit, quit, or empty, exit silently.
   • If the input is invalid, print Not a valid selection. and exit.

6. For partial reports. Files matching *-PARTIAL.md (incomplete runs) appear in the listing with a ⚠️ prefix in the Verdict column to make it clear they don't represent a complete pipeline run.

Edge cases

• No .review/ directory. Treat as no past reviews.
• A report file exists but has no parseable metadata. Show it with ? placeholders. Don't crash.
• Many reports (≥ 50). Limit the table to the most recent 25 entries. Print a footer: ... and <N> older reviews. Browse .review/reports/ directly to see all.
• The user pipes input or runs in a non-interactive shell. If stdin is not a TTY, skip the selection prompt and only print the table.

Notes

This skill is read-only. It does not modify any files, does not invoke any subagents, and does not consume model budget. It's purely a navigator over the user's review history.