Content
75%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 well-executed operational skill: copy-paste commands for every core operation, path conventions treated as the source of truth, and honest edge-case handling (soft-close semantics, goal_id rename rejection). The main gaps are the under-specified edit operation, the absence of any post-write verification step, and some repetition around the path-encoding rule.
Suggestions
Add a concrete example to "Edit a goal description" (e.g. a Read/Edit sequence on an actual goal path) so every operation has the same level of executable detail.
Add a verification step after create/capture (e.g. re-run `hivemind goal list --mine` or `ls` on the status folder to confirm the row landed) to close the validation gap in the workflows.
State the path-encoding rule once (in Path conventions or Constraints) and reference it elsewhere instead of repeating it three times, and consider moving the parked-task capture/resume procedures (sections 1a/1b) into a reference file to slim the main body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dominated by commands, path templates, and tight examples rather than concept explanations, matching the anchor for efficient content with minor trims available. It is not 5 because the path-is-source-of-truth rule is stated three times, the `mv` command is repeated nearly verbatim, and emphasis caps ("LEARN THESE", "DO NOT") pad a few sections. | 4 / 5 |
Actionability | Nearly every operation ships a copy-paste command (`ls`, `mv`, `rm`, `hivemind goal add/get/list`, `uuidgen`) plus a fully worked capture example, but section 3 ("Edit a goal description") contains only a comment block with no concrete command or example. This matches the anchor for mostly executable guidance with minor gaps; not 5 because editing a goal is a common case left without a worked example. | 4 / 5 |
Workflow Clarity | Create and resume flows are numbered with dependencies resolved up front (`hivemind whoami` → uuidgen → write) and include an ambiguity checkpoint ("If ambiguous, show the candidates and ask"); `rm` is explicitly defused as a soft-close so the destructive-op cap does not apply. Matches the anchor for clear sequence with minor validation gaps — no post-write verification step (e.g. re-list to confirm the row landed); not 3 because sequences and some checkpoints are explicitly present. | 4 / 5 |
Progressive Disclosure | No bundle files exist and the skill is self-contained, with clean sections (When to use / Path conventions / File body format / Operations / Constraints / Team visibility) and no buried or nested references. Matches the anchor for good structure with minor organization gaps; not 5 because at ~136 lines it exceeds the simple-skill case and inlines two related but separable procedures (goals vs. parked-task capture/resume) that could live in a reference file. | 4 / 5 |
Total | 16 / 20 Passed |