Content
67%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-organized behavioral skill that conveys its stance effectively with concrete anchors (a runnable command, real artifact paths, and a specific insight-to-artifact capture table). Its main weakness is redundancy: the stance principles are repeated across three sections and the illustrative transcripts are long, inflating token cost without adding proportional guidance.
Suggestions
Merge 'The Stance', 'What You Don't Have To Do', and 'Guardrails' into a single section - they restate the same principles ('Don't rush', 'Don't auto-capture', 'Do visualize', 'Don't force structure') nearly verbatim.
Trim the four entry-point transcripts to one or two representative examples, or move them to a separate examples reference file, keeping only the ASCII-diagram variety inline.
Add the concrete command for transitioning out of explore mode (e.g., creating a change proposal) so the 'exit explore mode first' reminder is executable rather than a pointer.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body restates its core principles three times across 'The Stance' ('Do visualize', 'Grounded'), 'What You Don't Have To Do' ('Don't rush', 'Be brief'), and 'Guardrails' ('Do visualize - A good diagram is worth many paragraphs', 'Don't rush', 'Don't auto-capture'), and the four entry-point transcripts occupy roughly a third of the file. It never explains concepts Claude already knows, so it stays above anchor 2, but the duplicated stance sections and long illustrative dialogs could be tightened, matching anchor 3. | 3 / 5 |
Actionability | For an instruction-only skill the guidance is largely concrete: the command 'openspec list --json', specific artifact paths ('openspec/changes/<name>/proposal.md', 'design.md', 'tasks.md'), a table mapping insight types to capture files ('specs/<capability>/spec.md'), and modeled offer phrasings ('This feels solid enough to start a change. Want me to create a proposal?'). It falls short of anchor 5 only because the core exploration behavior is conveyed through illustrative transcripts rather than directly executable instruction, matching anchor 4. | 4 / 5 |
Workflow Clarity | The 'When a change exists' branch is an explicitly numbered sequence (read artifacts -> reference naturally -> offer to capture -> 'The user decides') with a clear checkpoint against auto-capturing, and the single concrete command 'openspec list --json' anchors the context check. It falls short of anchor 5 because the mode is intentionally unsequenced and minor gaps remain (e.g., no explicit command for exiting explore mode to create a proposal), matching anchor 4's 'clear sequence with most checkpoints present'. | 4 / 5 |
Progressive Disclosure | As a single-file skill with no bundle directories, the content is well organized under clear section headers (Stance, What You Might Do, OpenSpec Awareness, Entry Points, Ending, Guardrails) with no nested references. It falls short of anchor 5 because the four long entry-point dialogue transcripts (~100 lines of illustrative content) could live in a separate examples file rather than the main SKILL.md, matching anchor 4's 'minor organization gaps'. | 4 / 5 |
Total | 15 / 20 Passed |