Content
77%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 exceptionally actionable and workflow-safe: exact commands, documented script I/O, exit-code recovery loops, and honest-degradation rules for sparse data. Its weaknesses are redundancy in the Gotchas section — thresholds restated from Phase 2b and a long-transcript counting gotcha that contradicts the script mandate — and a referenced helper script that is absent from the provided bundle.
Suggestions
Deduplicate the Gotchas section: remove 'Pattern mode needs at least 5 meetings' and 'Outcome correlations need at least 3 per group' (Phase 2b states both), and delete or rewrite the 'Long-transcript accuracy degrades' gotcha, which contradicts the instruction to compute all counts via mirror_metrics.py.
Resolve the broken bundle reference: scripts/mirror_metrics.py is referenced by the core pipeline command but does not exist in the provided bundle — add it, or inline the metric logic and drop the external path.
Tighten command consistency: pass --json to 'minutes list' in both modes (Phase 2b says 'filter its JSON results' but the command omits the flag), and specify the actual 'bounded search' command for locating a user-named meeting.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient — commands, the metric-field table, and both output templates carry operational weight — but the Gotchas section restates the '>=5 meetings' and '>=3 per tag group' thresholds already specified in Phase 2b, and the 'Long-transcript accuracy degrades' gotcha ('sample three 1500-word segments and extrapolate') directly contradicts the mandate to 'Compute the metrics with the bundled helper script, not by counting in-context'. This is 'some unnecessary explanation or could be tightened' rather than the merely trimmable minor instances of anchor 4. | 3 / 5 |
Actionability | Fully executable throughout: exact commands ('minutes voices --json', 'cat ~/.minutes/config/self.txt', 'mkdir -p ~/.minutes/config; printf …'), the complete 'set -o pipefail' pipeline into mirror_metrics.py with the '--self' flag, a field-by-field table of script output, exit-code semantics with a concrete recovery path ('On exit code 3 … re-run with one of those, or update ~/.minutes/config/self.txt'), and copy-paste output templates. This matches the copy-paste-ready anchor; no gaps for the common single-meeting and pattern cases. | 5 / 5 |
Workflow Clarity | Five clearly sequenced phases (0-3) with explicit validation ('Require exit status 0', 'Require both sides of the pipeline to exit successfully'), genuine feedback loops (exit code 3 → re-run with the discovered speaker labels; ask-once-and-cache in Phase 0), and degradation checkpoints ('Baseline computed from only N meetings, treat with caution', 'If every result is null, skip the outcome-correlation section'). This is the validate→fix→retry anchor fully realized; 4 would apply only if recovery guidance were missing. | 5 / 5 |
Progressive Disclosure | In-file structure is clean (phased headers, tables, templates) and the single external reference — ${CLAUDE_PLUGIN_ROOT}/skills/minutes-mirror/scripts/mirror_metrics.py — is clearly signaled and only one level deep. However, no scripts/, references/, or assets/ directories exist in the provided bundle, so the skill's central dependency cannot be verified to exist and the core pipeline command is unresolvable; that broken primary reference prevents the 'minor organization gaps' fit of anchor 4. | 3 / 5 |
Total | 16 / 20 Passed |