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.
An exceptionally actionable, well-sequenced orchestration skill — the tool-call specs, round templates, and error-recovery paths leave nothing to interpretation. Its costs are payload weight: ~450 inline lines where the member prompts and round templates belong in reference files, plus skeleton-repetition across the five prompts and triple-statement of the planner-handoff contract.
Suggestions
Move the five member system prompts to references/ (e.g., references/members/skeptic.md) and keep only the role/attack-vector table inline, cutting ~140 lines from the always-loaded context.
Factor the shared prompt skeleton (HOSTILE list, ruthlessness clause, default-position-on-others'-findings, output format) into one stated convention, then give each member only its unique role and weapons.
State the Phase 6 handoff contract once (Phase 6 itself) and reduce the three overlapping anti-pattern rows to a single row referencing it.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient skill-specific instruction, but the five member system prompts (~140 lines) repeat an identical skeleton ("You are HOSTILE to...", "Be ruthless...", "When you receive others' findings, default position...") and the Phase 6 handoff contract is restated in the Phase 5 intro, Phase 6 itself, and three anti-pattern rows. This is tightening-eligible redundancy that could be trimmed without losing the mechanism — anchor 3 rather than anchor 2, since nothing explains concepts Claude already knows and the padding is emphatic rather than filler. | 3 / 5 |
Actionability | Every step ships executable, copy-paste-ready material: the full `team_create` inline spec with all five member entries, verbatim round-task message templates for Rounds 1-3, the exact insight-bundle markdown shape, the complete `task` planner-dispatch call with the full handoff prompt, and the exact `team_delete` cleanup call. Placeholders are explicitly marked ("<full Skeptic system prompt>", "[insert Round 1 Findings Bundle]") — this matches anchor 5 fully. | 5 / 5 |
Workflow Clarity | The 7 phases are strictly sequenced with explicit collection gates ("the round is complete once every expected reply has arrived"), precondition verification (tools present, lead-only session), a degradation path (deep-low rejection → retry without researcher, state degraded roster), and feedback loops for failure recovery (nudge silent members once, surface team_delete failure with a manual retry, forward planner clarifying questions). Anchor 5's 'explicit validation steps; feedback loops for error recovery; checklists' is met by the anti-patterns table and per-phase gates. | 5 / 5 |
Progressive Disclosure | Section structure and navigation within the file are good, but this is a ~450-line monolith with zero bundle files: the five member system prompts and the three round-task templates are exactly the content that belongs in per-member reference files (the body even says 'substitute each prompt with the full system prompt above', signaling they are separable payloads). Anchor 3 (structure present, content that should be separate is inline, no clear references) fits; anchor 2 is too harsh since headers and tables keep it navigable, but the 50-line simple-skill exemption clearly does not apply. | 3 / 5 |
Total | 16 / 20 Passed |