Content
70%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 a well-gated, disciplined debugging workflow with unusually strong validation checkpoints and error-recovery feedback loops. Its weaknesses are structural: a large inline activation script with redundant prose re-explanation inflates token cost and makes the file monolithic rather than progressively disclosed.
Suggestions
Move the freeze-activation bash block into a scripts/ file (e.g. scripts/freeze.sh) and invoke it in one line; the ~35 lines of code plus the paragraph re-describing its behavior are the body's largest token cost.
Delete or compress the prose that restates what the preceding script already shows (the Python 3 requirement, symlink/empty-state refusal, exclusive-creation details are all readable from the code).
Replace the <module-directory> placeholder with one concrete resolved example so the activation snippet is copy-paste ready as-is.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly dense, non-obvious instruction (3-Strike Rule, WTF scoring formula, freeze semantics), but the paragraphs following the freeze script re-explain what the just-shown code does ("Activation uses the shared host session resolver and requires a stable session ID", "Both activation and freeze enforcement require Python 3", "Existing empty, symlinked or nonregular state is refused"). That is more than minor over-explanation, so anchor 3 fits better than 4. | 3 / 5 |
Actionability | Guidance is mostly executable: a concrete loop (reproduce, retain failure signature, minimize, test one named hypothesis, verify both scenarios), quantitative rules (+15% per revert, +20% unrelated files, 20% stop threshold), a formatted example output line, and a copy-paste bash block with only a <module-directory> placeholder. Not anchor 5 because the core investigation steps are prose-directed rather than turnkey commands, and the referenced skills/blocks/*.md dependencies live outside this bundle. | 4 / 5 |
Workflow Clarity | Clear sequence with explicit validation checkpoints: the HARD-GATE blocks behavior changes before reproduction and a named hypothesis is tested, both the minimal reproduction and the original scenario must be verified after the fix, and "Confidence is not verification" forces re-running evidence. Feedback loops for error recovery are explicit (3-Strike Rule, mandatory strategy rotation after two failures, inconclusive exit path for missing dependencies). Matches anchor 5. | 5 / 5 |
Progressive Disclosure | The skill is a single ~115-line file with a ~35-line inline bash block plus explanatory prose that clearly belongs in a scripts/ or references/ file; no bundle files exist to offload it. External references are clearly signaled ("Read and apply skills/blocks/debug-feedback-loop.md"), which keeps it above anchor 2, but the structure is monolithic rather than well split — anchor 3. | 3 / 5 |
Total | 15 / 20 Passed |