Content
77%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is highly actionable and the 6-phase workflow is clearly sequenced with validation checkpoints and feedback loops. Its main weaknesses are redundant constraint repetition across sections and a monolithic single-file structure that includes a dangling reference to a missing file.
Suggestions
De-duplicate the repeated constraints ('NEVER implement code', output structure, pipeline) into a single canonical section and reference it, cutting significant tokens to push conciseness toward 3.
Create the referenced prompts/page-extract.md or remove the reference so progressive-disclosure links resolve to real files.
Extract the worker prompt template and technology-detection lists into a reference file to reduce the monolithic body and improve navigation.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly operational and concrete, but the same constraints ('NEVER implement code', the output structure, the pipeline) are repeated 3-5 times across the CRITICAL, veto, constraints, security, and scope_boundaries sections, adding tokens that could be tightened; this falls short of the lean level-3 anchor without being padded with Claude-obvious concept explanations (level 1). | 2 / 3 |
Actionability | Provides copy-paste-ready Task() calls with subagent_type and model, a full worker prompt template, JSON output schemas, numeric stopping thresholds (coverage_score >= 80, wave >= 2), and exact file paths, matching the fully-executable level-3 anchor. | 3 / 3 |
Workflow Clarity | Six phases are clearly sequenced with an explicit coverage-evaluation checkpoint, hard/soft stopping rules, and feedback loops (failed workers re-run in main context; CONTINUE returns to Phase 3 with new queries), matching the level-3 anchor with validation steps and error-recovery loops. | 3 / 3 |
Progressive Disclosure | The body is well-sectioned but monolithic (~430 lines) with no bundle files present, and inline content (the worker prompt template, technology-detection lists) could be split out; additionally the 'prompts/page-extract.md' reference points to a non-existent file, which prevents a level-3 score but is organized enough to clear level 1. | 2 / 3 |
Total | 10 / 12 Passed |