Content
88%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.
An excellent operational skill: copy-paste commands in every flow, a clearly sequenced upgrade procedure with validation and failure handling, and a tight security contract (hardcoded action, never parse commands from the marker). Remaining improvements are marginal — version-sensitive playbook pointers and possibly splitting the always-on runbook into a reference.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and operational throughout — no concept explanations, every section carries gbrain-specific facts Claude cannot know ("once per 30 minutes per version", snooze escalation "24h → 48h → 7d", exit-code semantics). The deduction is for time-sensitive version-specific material cited inline ("skills/migrations/v0.46.3.0.md", "#5038", "v0.50.0.0.md") rather than placed in an old-patterns/deprecated section, per the guidelines. | 4 / 5 |
Actionability | Fully executable: every flow is driven by copy-paste commands — `gbrain config get self_upgrade.mode`, `gbrain self-upgrade --check-only --json`, `gbrain autopilot pause --reason "upgrade"`, `gbrain upgrade --restart-only --dry-run --json`, `gbrain doctor --only runtime_build_skew,serve_build_drift --json` — plus a ready-to-present 4-option question template and exact one-line report strings for each outcome. | 5 / 5 |
Workflow Clarity | The always-on flow is a numbered 6-step sequence (pause → drain → preview → upgrade → verify → resume) with explicit validation checkpoints (dry-run preview in step 3, per-install `gbrain --version` and doctor checks in step 5) and error-recovery feedback loops (`lock_busy` wait-and-retry, `restart_refused` exit 3 relay, `failed_versions` no-retry). The notify flow confirms a real update via check-only before prompting, and the mutating/batch nature is covered by validation, so no cap applies. | 5 / 5 |
Progressive Disclosure | A single-file skill (no bundle files) with well-organized sections and one-level-deep, clearly signaled external references (the `skills/migrations/` playbooks with a published-repo link, `../conventions/agent-operator-protocol.md`, the upgrades guide). It sits below the anchor-5 example because nothing is split across files — the ~30-line always-on host runbook is inlined and could live in a reference file. | 4 / 5 |
Total | 18 / 20 Passed |