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.
A dense, highly prescriptive multi-phase workflow that is fully executable: exact commands, verbatim question scripts, decision tables for routing and rigor inference, and validation checkpoints (read-back checks, contradiction re-asks, non-zero helper handling) at every risky point. The only room for improvement is trimming the restated never-run-the-next-step rule and possibly splitting the Phase 6 path listings into a reference file.
Suggestions
State the "never run the next step yourself, the user types it" rule once (e.g. in the intro) and refer back to it from Phases 6 and 7 instead of repeating it three times.
Consider moving the per-rigor path listings in Phase 6 to a reference document to keep the interactive body focused on the dialogue flow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every sentence is a project-specific rule (e.g. "a written value pins that knob over the rigor expansion, and a mirror file outranks it", "`source_files` of `unknown` means the code root could not be found: say that, never 'no code'") with no padding or explanation of concepts Claude already knows. Not score 5 because the "never run the next step" rule is stated three times (intro "It never runs that step itself", Phase 6 "never to running the step", Phase 7 "Do not start that skill") and a few dense passages (the legacy mirror paragraph) could be tightened; well above score 3, which requires unnecessary explanation. | 4 / 5 |
Actionability | Fully executable throughout: exact commands ("bash .claude/scripts/project-state.sh", "bash .claude/scripts/set-stage.sh \"[stage]\"", "bash .claude/scripts/config-set.sh modes.rigor [value]"), verbatim question texts with option lists ("Welcome! Where are you starting from with your game?"), a copy-ready response card, and an exact one-line handoff ("Type `[skill]` to begin."). This matches anchor 5: copy-paste ready commands covering the common cases. | 5 / 5 |
Workflow Clarity | Seven clearly sequenced phases with explicit checkpoints and error-recovery loops: wait for the answer before proceeding, re-ask on contradiction ("If it contradicts Phase 1, say what you found once and ask again"), verify saves ("Check the read-back rows match"), and handle helper failure ("If a helper exits non-zero, show its output, say which setting was not written, and go on with the value for this run"). This matches anchor 5: clear sequence, explicit validation, and feedback loops for recovery. | 5 / 5 |
Progressive Disclosure | Well-organized phase sections with clearly signaled one-level references to project docs ("`.claude/docs/settings-guidance.md` §3", "per `.claude/docs/automation-modes.md`") and no buried or nested references; no bundle files exist to verify beyond the body itself. Not score 5 because everything lives in one ~210-line file — the per-rigor path listings in Phase 6 are a plausible candidate for a separate reference — though all of it is needed during the single interactive run. | 4 / 5 |
Total | 18 / 20 Passed |