orchestrix-skills 0.1.0

package/skills/review-code/SKILL.md

@@ -0,0 +1,71 @@
+---
+name: review-code
+description: Use after implementing a task or feature and before merging, to check a diff against its spec and for quality.
+license: MIT
+allowed-tools: [Read, Bash, Grep, Glob]
+metadata:
+  contract:
+    inputs: [diff, spec]
+    reads: [taste/coding-standards]
+    outputs: [review_report]
+    authority: "Read-only on code. Produces a report. No edits, no commits, no production."
+    verify: "Report contains both verdicts: spec compliance AND code quality."
+    accept:
+      when: "never — findings route back to implement; the human reviews at the end batch."
+      timing: deferred
+---
+
+# Review Code
+
+Review a diff against its spec with fresh eyes. Two verdicts, in order: does it
+do what was asked, then is it well built.
+
+**Core principle:** Review the work product against the spec — not the author's
+reasoning. You get the diff and the spec, not the session that produced them.
+
+## Inputs
+
+- `diff` — the change to review (a base..head range or a diff file).
+- `spec` — the story / acceptance criteria the change must satisfy.
+
+Read the spec first. It is your attention lens. Then read the full diff.
+
+## Verdict 1 — Spec compliance
+
+For each acceptance criterion: met, missing, or extra (built but not asked for).
+
+- **Missing** a criterion → spec verdict is FAIL.
+- **Extra** scope not in the spec → flag it; over-building is a defect.
+
+## Verdict 2 — Code quality
+
+Check, against `taste/coding-standards`:
+
+- Correctness and edge cases the tests miss
+- Duplication, unclear names, dead code
+- Tests that assert nothing or test mocks instead of behavior
+- Security / data-loss / trust-boundary issues (always Critical)
+
+Rate each finding: **Critical** (must fix) · **Important** (fix before merge) ·
+**Minor** (note it).
+
+## Output: `review_report`
+
+```yaml
+spec_compliance: pass # pass | fail
+missing: [<AC not met>]
+extra: [<scope not requested>]
+findings:
+  - { severity: Important, where: src/x.ts:40, issue: "...", fix: "..." }
+verdict: changes_requested # approved | changes_requested
+```
+
+## Rules
+
+- **Do not pre-judge.** Never decide a finding is fine because "the spec said
+  so" — surface it; conflicts with the spec are the human's call, not yours.
+- **Be specific.** Every finding has a location and a concrete fix.
+- **Findings route back, not to the human.** Critical/Important findings go to
+  `implement` (re-run with this report as `qa_feedback`). The human sees the
+  result at the end-of-run acceptance, not each review.
+- If the diff is clean, say so plainly and approve.

package/skills/run-tests/SKILL.md

@@ -0,0 +1,73 @@
+---
+name: run-tests
+description: Use before claiming any work is complete, fixed, or passing, and before committing or handing off.
+license: MIT
+allowed-tools: [Read, Bash]
+metadata:
+  contract:
+    inputs: [target, expected_outcome?]
+    reads: []
+    outputs: [verification_report]
+    authority: "Run test, lint, and build commands. Read-only on source. No edits, no production."
+    verify: "self — this skill IS the objective verification primitive."
+    accept:
+      when: "never"
+      timing: deferred
+---
+
+# Run Tests (Verify Before Completion)
+
+Claiming work is done without running the check is a false claim, not a shortcut.
+
+**Core principle:** Evidence before claims, always. If you did not run the
+command in this step, you cannot say it passes.
+
+## The Iron Law
+
+```
+NO COMPLETION CLAIM WITHOUT FRESH VERIFICATION EVIDENCE
+```
+
+## The gate
+
+```
+1. IDENTIFY the command that proves the claim.
+2. RUN it fully and fresh (not a remembered run).
+3. READ the full output: exit code, pass/fail counts, warnings.
+4. DECIDE: output confirms the claim, or it does not.
+5. REPORT the claim WITH the evidence.
+```
+
+## Verify independently — do not trust self-reports
+
+| Claim                   | Proof required                      | Not sufficient                |
+| ----------------------- | ----------------------------------- | ----------------------------- |
+| Tests pass              | Test command output: 0 failures     | "should pass", a previous run |
+| Build succeeds          | Build command: exit 0               | linter passed                 |
+| Bug fixed               | Test of the original symptom passes | code changed, assumed fixed   |
+| Another skill succeeded | Its diff and tests checked here     | its success report            |
+
+A regression test counts only after a red→green check: revert the fix, watch
+the test fail, restore the fix, watch it pass.
+
+## Output: `verification_report`
+
+```yaml
+target: <what was checked>
+command: <exact command run>
+exit_code: 0
+result: { passed: 34, failed: 0, warnings: 0 }
+verdict: pass # pass | fail
+evidence: |
+  <relevant lines of real output>
+```
+
+`verdict: fail` is a valid, honest result. Report it with evidence; never round
+a failure up to a pass.
+
+## Red flags — stop
+
+- "should", "probably", "looks correct", "Done!" before running the command
+- Reusing an earlier run's result
+- Partial check used to claim full success
+- Trusting "the agent said it worked"