Content
60%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 information-rich, highly actionable tip catalog with excellent per-tip commands and a helpful quick-reference table, but it is structurally weak as a skill: everything is crammed into one monolithic file with no progressive disclosure, and content is diluted by duplicated sections, provenance meta-listing, and time-sensitive version claims.
Suggestions
Split the 60 sections into per-topic reference files (e.g. references/worktrees.md, references/hooks.md, references/scheduled-tasks.md) and keep SKILL.md as a concise overview with the Quick Reference table linking to them — this addresses the progressive_disclosure score directly.
De-duplicate the repeated sections (/btw §33/§54, /batch §30/§56, /loop §31/§48, worktrees §1/§28/§55, hooks §7/§24/§49) and remove the 9-thread 'Parts' provenance block, which is meta-content about sourcing rather than actionable guidance.
Move time-sensitive details (thread dates, version gates like 'v2.1.74+', rollout percentages) into a clearly labeled 'versions & history' section so stale claims don't pollute the evergreen tips.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Individual tips are terse and command-dense ("git worktree add .Codex/worktrees/my-worktree origin/main", "/loop 5m /babysit"), but the body carries real bloat: a 9-thread "Parts" provenance block, near-verbatim duplicate sections (/btw in §33 and §54, /batch in §30 and §56, /loop in §31 and §48, worktrees in §1/§28/§55, hooks in §7/§24/§49), and unquarantined time-sensitive claims ("Jan 2, 2026", "v2.1.74+", "rolled out to 100% of users"). Most sections are unique and efficient, so this sits between anchors 2 and 3 — the duplication exceeds 'minor trimming' but does not dominate the file. | 3 / 5 |
Actionability | Strong concrete guidance throughout: executable bash ("Codex --worktree my_worktree --tmux", "Codex -p ... --bare"), complete JSON/YAML blocks (PostToolUse hook config, agent frontmatter), and exact prompt phrasing ("Grill me on these changes and don't make a PR until I pass your test"). Not a 5 because a meaningful share of tips remain hand-wavy ("Ask Codex to add a hook to get started", "Build a spaced-repetition learning skill"). | 4 / 5 |
Workflow Clarity | As a lookup catalog, the numbered sections plus the Quick Reference table make finding and applying the right tip unambiguous, and per-topic mini-workflows (create worktree → launch session → alias hopping) are correctly sequenced with no destructive or batch operations that would require validation checkpoints. Not a 5 because there is no error-recovery or feedback-loop guidance, and riskier suggestions (auto mode, "--dangerously-skip-permissions") are presented without caveats. | 4 / 5 |
Progressive Disclosure | A ~895-line monolithic SKILL.md inlines all 60 topics with zero bundle files ("All tips are contained in this file — do not fetch from the website") — content that clearly belongs in separate per-topic reference files. Section headers and the Quick Reference table prevent a score of 1, but with no references at all and a volume far beyond the anchor-3 example's 200 inline lines, the inlining problem dominates. | 2 / 5 |
Total | 13 / 20 Passed |