Content
85%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is an exemplary instruction-only guard: exact flags, commands, env vars, and a verbatim announce format, with a clearly sequenced workflow containing validation checkpoints and a diagnosis feedback loop. Its one weakness is token efficiency — the Rules section substantially duplicates the Workflow steps and could be collapsed into the invariants that add new information.
Suggestions
Collapse Rules bullets that restate Workflow steps 3–6 verbatim (MCP flag split, read-only, wrong-deployment) and keep only the invariants that add new framing ("per-action, per-target, per-session") or new information (the composition rule naming ship, env, migrate, seed).
Trim the motivational intro ("Deployments are not interchangeable, and most incidents start with a command aimed at the wrong one") to a single clause, since Claude can infer the motivation from the guard's existence.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is telegraphic and dense, but the Rules section restates Workflow steps 3–6 nearly verbatim (e.g., the MCP-flag rule duplicates step 4, the read-only rule duplicates step 5), roughly 30% duplicated content — matching anchor 3 ("could be tightened") rather than anchor 4's "minor instances of over-explanation". | 3 / 5 |
Actionability | Fully concrete, executable guidance: exact env vars and files to read (CONVEX_DEPLOYMENT in .env.local, convex.json, CONVEX_DEPLOY_KEY), exact commands (npx convex run --prod), exact MCP flags (--cautiously-allow-production-pii, --disable-tools run,envSet,envRemove), and a verbatim announcement format ("target: dev (joyful-capybara-123, personal dev)") — a match for anchor 5. | 5 / 5 |
Workflow Clarity | The 7-step sequence has explicit validation checkpoints ("If two sources disagree, resolve before proceeding", fresh-consent gate, "Ambiguity = stop ... never guess") plus an error-recovery feedback loop ("do NOT re-deploy harder. Re-run step 1"), matching anchor 5; the destructive-operation cap does not apply because validation is present. | 5 / 5 |
Progressive Disclosure | The skill is under 50 lines with no bundle files and no need for external references, and its sections (Workflow, Rules) are well-organized — per the rubric's simple-skill guideline this scores 5 on structure alone. | 5 / 5 |
Total | 18 / 20 Passed |