Content
32%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 core instructions establish a sensible philosophy (3-way merge reasoning, HITL on uncertainty) but never give an executable, ordered procedure with validation checkpoints, while the bulk of the file is an ever-growing lessons log that should be a separate reference. Routing a merge through this skill would work, but the agent would have to reconstruct the workflow from scattered prose and buried commands.
Suggestions
Add a numbered workflow with concrete commands and validation checkpoints, e.g.: 1. `git fetch origin main` 2. Diff each side against merge-base (`git diff --name-only <base> origin/main` vs `… HEAD`, `comm -12` for overlaps) 3. `git merge origin/main` 4. Resolve conflicts per policy, HITL if not 100% sure 5. **Validate**: run both `src/rosettify-plugins` and `src/hooks` test suites, regenerate plugins, re-diff — fix and re-validate on failure 6. Report.
Move the lessons log to `references/lessons.md` (or a curated top-5 distilled into short imperatives) and keep only the template plus a one-line pointer in SKILL.md, restoring progressive disclosure and bounding token cost.
State the conflict-resolution policy concretely (what '3-way merge logic' means operationally: reconstruct the common ancestor, compose both intents per-hunk, prefer the shape that can derive the other) instead of relying on 'Not extremes. Not mechanical.'
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The instruction core (persona, task, conflict policy, HITL rule, workspace map) is lean, but the '## Lessons learned' section has accumulated ~16 entries of 80–150-word dense narrative — a noticeably verbose padded log inlined where a distilled set belongs. Not a 1 because the lessons are repo-specific knowledge Claude genuinely lacks, and not a 3 because the section's unbounded 'keep updating' design guarantees further padding. | 2 / 5 |
Actionability | The primary workflow gives abstract direction only — "Sync+merge remote origin main", "Use 3 way merge logic. Not just selecting A vs B" — with no actual git commands for fetch/merge/abort. Concrete, executable commands do appear ("git diff --name-only <base> origin/main", "comm -12 …", "diff -rq plugins/<t> plugins/<t>-light", running both src test suites) but only buried inside lessons, leaving the main flow incomplete. Fits anchor 3: 'Some concrete guidance but incomplete; missing key details'. | 3 / 5 |
Workflow Clarity | Merging is a risky, potentially destructive operation, yet the steps are scattered in prose with no ordered sequence (no fetch → overlap-diff → merge → resolve → verify → report pipeline) and validation exists only implicitly inside lessons ("always run both src/rosettify-plugins and src/hooks suites"). The rubric's cap of 3 for missing validation applies, and the content fits anchor 2 better: 'Rough sequence present but many gaps; steps poorly defined; validation absent [as explicit checkpoints]'. | 2 / 5 |
Progressive Disclosure | The body has a single section header, and the ~8 KB 'Lessons learned' log is exactly the case of 'content that clearly belongs in separate files is inlined' — it is reference material (per-incident merge case studies) that should live in a references/ file with the SKILL.md keeping the template and a one-line pointer. Not a 1 because there is some structure (the lessons section is delimited and templated) and references are not nested. | 2 / 5 |
Total | 9 / 20 Passed |