Content
57%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-sequenced two-phase protocol with a strong user-confirmation gate and much copy-paste-ready detail (commands, verbatim prompts, directory structure, .gitattributes content). Its weaknesses are repetition used as emphasis, delegation of the casting algorithm to a reference that cannot exist yet at load time, inlined project-specific governance policy, and no verification step after the batch file creation in Phase 2.
Suggestions
Add a post-creation verification step at the end of Phase 2 (e.g., confirm team.md contains the exact '## Members' header and the casting registry/history were initialized by the locked SDK writer) before announcing '✅ Team hired', closing the feedback loop on the batch scaffolding operation.
Fix the circular reference in Phase 1 step 3: '.squad/templates/casting-reference.md' cannot exist when Init Mode triggers (no .squad/ yet). Either inline a minimal casting algorithm or point to a file actually shipped with this skill's bundle.
Move the three Waza-specific governance override paragraphs (casting-state repair and reviewer seeding) into a separate reference file, and collapse the repeated roster-copy rules and the four per-agent 'exempt from casting' lines into single statements.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is operational rather than explanatory, but repeats itself noticeably: the roster-copy requirement is stated three ways ("Do not regenerate, summarize, truncate, omit, or replace any roster row", the shape block, then "Never use a bare 'Look right?' question"), "genuinely empty new target" appears twice nearly verbatim, and the four always-on agents each get a separate "exempt from casting" line that could be one sentence. This is mostly-efficient-but-could-be-tightened rather than anchor 2's padded sections, since there is no explanation of concepts Claude already knows beyond one justifiable line about the union merge driver. | 3 / 5 |
Actionability | Most guidance is copy-paste ready: exact command ("git config user.name"), verbatim questions, the full .squad/ directory list, the complete .gitattributes block, the exact choices array, and the hard-coded "## Members" header requirement. The gap is that the core casting step delegates to "the canonical Casting reference at .squad/templates/casting-reference.md" — a file that cannot exist during Phase 1 since .squad/ is only created in Phase 2 — and seeding sources ("Rai-charter.md", "rai-policy.md", "fact-checker-charter.md") are named without paths. | 4 / 5 |
Workflow Clarity | Sequencing is excellent — numbered steps across two phases, a hard "⚠️ STOP" gate before any file creation, an explicit Phase 2 trigger, and a re-propose feedback loop for "add someone"/"change a role". However, Phase 2 performs a batch operation (creating the whole .squad/ tree plus updating .gitattributes) with no post-creation verification step — e.g., no check that team.md actually contains the required "## Members" header or that casting state initialized atomically — so the rubric's cap for batch operations missing validation/verification applies. | 3 / 5 |
Progressive Disclosure | The two-phase structure with headers and numbered steps is clear, but three dense Waza governance-override paragraphs (casting-state repair, reviewer seeding) are inlined policy that belongs in a separate reference file, and the references the body does cite (.squad/templates/casting-reference.md, squad.agent.md sections, charter templates) are neither bundle files nor clearly resolvable at load time. This matches the anchor for structure present but content that should be separate kept inline with references not clearly signaled. | 3 / 5 |
Total | 13 / 20 Passed |