Content
85%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.
A tight, well-sequenced workflow with genuine validation and feedback loops, and excellent token efficiency. The main weaknesses are the unreferenced bundle script that duplicates step 1's instructions and the correspondingly vague corpus-pull direction.
Suggestions
Reference the bundle script in step 1, e.g. "Run `python3 scripts/fetch_sent.py` (token from $VAULT_TOKEN_GMAIL_GOOGLEAPIS_COM) to build voice/corpus/corpus.jsonl", falling back to the selected email access skill only if it is unavailable.
If the script is the intended path, trim the inline re-description of quote/signature stripping and internal/external tagging that the script already performs.
State when to prefer the script's --query/--limit flags (e.g., limiting to recent mail for a refresh) so the bundle is navigable from SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~77-line body is lean and assumes competence: exact file names (voice/corpus/corpus.jsonl, email-voice-profile.md), thresholds ("fewer than ~50 usable emails"), and ready commands. The few explanatory asides ("Computers are replaced without notice...", "Gmail appends it on send") are single-line rationales for non-obvious rules, not padding of concepts Claude already knows, so it matches the 5-anchor rather than the 4-anchor's trimmable over-explanation. | 5 / 5 |
Actionability | Mostly executable: concrete API calls with headers, an exact curl upload command, the exact profile filename and section structure, and a defined corpus format. It falls short of 5 because step 1 says only "Use the selected email access skill to retrieve up to 300 of the user's sent messages" while the bundle's scripts/fetch_sent.py — which implements exactly that fetch — is never invoked, leaving the corpus pull as high-level direction where a single command exists. | 4 / 5 |
Workflow Clarity | Sections 0-5 give a clear sequence (check Files → pull corpus → study → write profile → validate → publish) with explicit validation steps and feedback loops: hold out 3 threads and tighten where drafts diverge, a small-corpus checkpoint that asks the user whether to continue, and user corrections folded in as hard rules. This matches the 5-anchor's explicit validation with error-recovery loops. | 5 / 5 |
Progressive Disclosure | The body is well-sectioned, but scored against the actual bundle: the one bundle file (scripts/fetch_sent.py) is never referenced or signaled anywhere in SKILL.md, so the skill file is not a navigable map of its own bundle. It fits the 3-anchor (structure present, references not clearly signaled) better than 4, since the sole provided script is undiscoverable from the overview. | 3 / 5 |
Total | 17 / 20 Passed |