Score

ABOUTME: Quality gate scoring: runs make check + make test-e2e and grades against quality-gates

ABOUTME: Reports commit-ready (>=80), PR-ready (>=90), excellence (>=95) with honest auto-fails

/score

Compute a quick quality readiness signal for the current working tree.

Process

1. Run make check && make test-e2e.
2. Fail = 0: if either target fails, score is 0 (auto-fail per quality-gates: tests/build broken is CRITICAL).
3. Pass = 100 baseline: if both pass, the static gate is green. For a full rubric-based score (Major/Minor deductions), invoke /gemini-review or /advanced-review and subtract:
   • 10 per Major finding
   • 3 per Minor finding
4. Report against the three thresholds from rules/quality-gates.md:
   • ≥80 commit-ready
   • ≥90 PR-ready
   • ≥95 excellence
5. Log the run, then render the trend (see History below). Never hand-write the row or the delta: the script computes both.

Output format

Score: <n>/100
Gate:  <commit-ready | PR-ready | excellence | BLOCKED>

Breakdown:
  make check:     <PASS|FAIL>
  make test-e2e:  <PASS|FAIL>
  Major findings: <n> (-<10n>)
  Minor findings: <n> (-<3n>)

Ready to commit:  <yes|no>
Ready to open PR: <yes|no>
Excellence:       <yes|no>

History

After the report, append the run and show where it sits against the previous ones:

SCORE_LOG="${CLAUDE_FORGE_ROOT:-$HOME/Development/private/claude-forge}/scripts/score-log.sh"
"$SCORE_LOG" --score <n> --threshold <t> --gate <commit|pr|excellence> --check <pass|fail> --e2e <pass|fail> --major <n> --minor <n>
"$SCORE_LOG" --trend

Paste the --trend output verbatim under the report. Values come from the breakdown just computed; --threshold and --gate are the pair the canonical trace line prints, SCORE: <n>/100 (threshold: <t>, gate: commit|pr|excellence) (rules/orchestrator-protocol.md): the action this run was aiming at and the number it is judged against, not the highest bar the score happened to clear. Log the same pair you print, or the row cannot be reconciled with its SCORE event.

• The script does the append and the arithmetic. The model supplies the seven measured values and nothing else: a hand-written row or a hand-computed delta is exactly the failure this indirection removes.
• The history file (quality_reports/score-history.jsonl at the target repo's git root) is a denormalized view of harness-trace SCORE events; change one, change the other.
• The script gitignores the file in the target repo on first write. It stays local: a branch name next to a low score corroborates gitignored review findings.
• Script missing or the run fails: report the score anyway and say the history was not written. A logging failure never blocks the gate.

Rules

• Be honest. If the output is ambiguous, report "inconclusive" and say why. Do not inflate.
• CRITICAL trumps all. A failing make check/test-e2e yields score 0 regardless of other factors.
• Missing gate = not zero. If make check or make test-e2e targets don't exist in the project, do NOT claim score 100. Report "gate missing, run /project-checks to scaffold" and refuse to score.
• Review integration is optional. If the user didn't ask for a full review, report the static gate result only and note that a rubric-based score requires a review agent pass.

When to use

• Before git commit: quick readiness check.
• Before opening a PR: confirm ≥90 threshold.
• Mid-session checkpoint: answer "are we done?" with a number instead of vibes.

When NOT to use

• Pre-commit already running as a hook: the hook gives a go/no-go; /score gives you a number and a breakdown. Different use.
• Large refactors where full review is warranted: go straight to /advanced-review, not /score.