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.
The body is a disciplined, fail-closed orchestration shim: it delegates the heavy workflow to explicitly-loaded external files and gives concrete dispatch and error-handling rules. Its main weakness is redundancy — the flag-activation logic and --wave documentation appear twice — and the absence of a literal gsd_dispatch example.
Suggestions
Deduplicate the flag rules: state the literal-token activation rule once (in <context>) and remove the 'Flag handling rule' bullets from <objective>, keeping only the one-line --wave mention.
Add a short concrete example of the dispatch contract, e.g. 'gsd_dispatch(role: implementer, task: <step prompt>)', and of the --wave flag syntax ('--wave 2').
Clarify what 'Preserve all workflow gates' concretely requires (checkpoint handling, verification, state updates are listed but not what each gate demands), or trust the referenced workflow file explicitly and drop the enumeration.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly lean and operational ('Orchestrator stays lean', 'Context budget: ~15% orchestrator, 100% fresh per subagent'), but the flag-activation rule is stated twice in near-verbatim form — the 'Flag handling rule' bullets in <objective> and the 'Active flags must be derived' bullets in <context>, with --wave documented in both sections. That duplication goes beyond the 'minor instances of over-explanation' of score 4, fitting 'mostly efficient but could be tightened'. | 3 / 5 |
Actionability | Guidance is concrete and executable for an instruction-only skill: exact tool contract ('call gsd_dispatch with role ... and task ...'), fail-closed error handling ('STOP that step and surface the error', offer to skip if optional), and a concrete query ('gsd-tools query init.execute-phase'). It falls short of 5 because no literal invocation example of gsd_dispatch or the wave-filter syntax is shown, leaving minor gaps. | 4 / 5 |
Workflow Clarity | The sequence is coherent — read execution_context files first ('Before doing anything else, read every file listed'), then execute the loaded workflow end-to-end — with explicit validation checkpoints for a batch operation: fail-closed dispatch handling and 'phase verification/completion still only happens when no incomplete plans remain'. Not score 5 because the actual step-by-step wave workflow is delegated to the external workflow file, so checkpoints like verification routing are named ('Preserve all workflow gates') rather than spelled out. | 4 / 5 |
Progressive Disclosure | The body is a well-sectioned overview (<objective>, <execution_context>, <context>, <process>) with clearly signaled one-level-deep references to the workflow and brand files, and no bundle files exist to nest. It misses 5 because flag documentation is inlined and duplicated across two sections rather than living once in the appropriate place, a minor organization gap. | 4 / 5 |
Total | 15 / 20 Passed |