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 exemplary operational skill: every command is executable, the workflow has explicit validation and adversarial self-checks, and it adds only repo-specific knowledge Claude could not infer. The only weaknesses are a small amount of duplicated command detail and the inlined runtime-fix harness that could live in its own file.
Suggestions
Deduplicate the heapsnapshot-delta.mjs invocation: state the command and flags once (Snapshot Comparison) and reference it from 'Validating runtime fixes' instead of repeating the block.
Consider moving the 'Validating runtime fixes' harness details (modes, snapshot output dir, fixture guidance) into a references/ file, keeping a two-line pointer plus the pnpm leak:embedded-run command in SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence — no explanation of what Vitest or heap snapshots are, and every line carries repo-specific evidence (config names, script paths, artifact files). It stops short of anchor 5 because the heapsnapshot-delta invocation is repeated nearly verbatim in both the 'Snapshot Comparison' and 'Validating runtime fixes' sections and could be trimmed. | 4 / 5 |
Actionability | Guidance is fully executable: exact pnpm commands with flags and output paths ('pnpm test:perf:groups --config test/vitest/vitest.unit-fast.config.ts --allow-failures --output ...'), a copy-paste ready /usr/bin/time incantation, and concrete script invocations with useful flags (--top 40, --min-kb 32, --pid 16133). These cover the common cases directly. | 5 / 5 |
Workflow Clarity | The five-step workflow is clearly sequenced (reproduce, collect evidence, classify, fix, verify) with an explicit verification step ('Re-run the same grouped or scoped command and confirm the max-RSS trend or OOM is reduced') and feedback loops — e.g., the closure-inline and synthetic-leak harness modes that validate the detector before trusting a PASS. Ambiguity handling ('stop short of a confident label and inspect retainers') adds an error-recovery path. | 5 / 5 |
Progressive Disclosure | The bundle's one script (scripts/heapsnapshot-delta.mjs) is well-signaled with usage shown inline, and the body is organized into clear sections (Workflow, Heuristics, Snapshot Comparison, runtime-fix harness, Output Expectations). It does not reach anchor 5 because the ~30-line 'Validating runtime fixes' harness section is a semi-separate concern inlined in SKILL.md, and the pointer to the sibling skill '../openclaw-test-performance/SKILL.md' is a cross-skill reference rather than a bundled one-level-deep reference. | 4 / 5 |
Total | 18 / 20 Passed |