Content
90%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 content is an exceptionally dense, fully executable operational specification: exact dispatch templates, exact commands, per-outcome handling with validation, re-dispatch caps, and security-conscious data handling throughout. Its only real weaknesses are readability — the compressed run-on paragraphs and heavy indirection to the external workflow file make the sequence effortful to follow — not missing substance.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is almost purely operational — routing rules, a dispatch task template, exact handling per return header, and commit commands — with zero explanation of concepts Claude already knows and no padding. Every sentence carries instruction weight (e.g. "Stop after 3 re-dispatches, or when two consecutive re-dispatches leave next_action equal to last_next_action"), matching anchor 5 ('Lean and efficient; assumes Claude's competence; every token earns its place'). | 5 / 5 |
Actionability | Guidance is fully executable: a literal copy-paste dispatch task template with placeholder semantics, exact commands ("git diff --cached --quiet || git commit -F <message file>", "gsd_run query commit: \"docs(debug): resolve <slug> session\" --files {debug_dir}/resolved/<slug>.md"), and exact first-match-wins handling for all six return headers plus the non-terminal and error cases. This matches anchor 5 ('Fully executable; copy-paste ready code or commands; specific examples cover the common cases'). | 5 / 5 |
Workflow Clarity | The loop is a clearly numbered sequence with real validation checkpoints and feedback loops — state-field sanitization for untrusted input, a 3-re-dispatch cap, no-progress detection via last_next_action, blocker reports, and per-header recovery options (e.g. 'Revise fix' re-dispatch after a guardrail rejection). It falls short of anchor 5 because the flow is hard to follow on first read: routing is spread across dense paragraphs, many steps ('run sections 1d, 2 and 3') defer to an external workflow file, and the compact-summary field list ('Session, Root Cause, Fix, Cycles, TDD...') is ambiguous. | 4 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ are all absent), and the single external dependency is one level deep and clearly signaled via the <execution_context> block pointing at gsd-core/workflows/debug.md. Structure is decent — bold section markers (**The loop.**, **Finish.**), numbered loop steps, a routing list — but it uses dense paragraphs rather than clean section headers, so navigation is good rather than easy, matching anchor 4 over anchor 5. | 4 / 5 |
Total | 18 / 20 Passed |