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.
The body is commendably lean and correctly pushes depth into two real, one-level-deep reference files, but its executable value is thin: the procedure stays abstract, no validation command is surfaced in the body, and a Spanish/mixed-language boilerplate section references bundle directories that do not exist. Formatting errors in the Contract section (collapsed nested bullets, a stray "- *.") further blur navigation.
Suggestions
Make the Procedure executable: name the actual commands and validation loop, e.g. "1. Scaffold: `python scripts/scaffold_spec.py --id <id> --agent <agent>` 2. Refine `Starter:` values 3. Validate: `python scripts/validate_spec.py <file>` until OK 4. Confirm with `scripts/check.sh`".
Fix or remove the "Packet" block: it lists knowledge/, prompts/, examples/, agents/, templates/ and a scripts/wire_packet.py that are absent from the bundle — align it with the real structure (references/, scripts/, assets/) so progressive disclosure matches reality.
Clean up the Contract section: translate or standardize the mixed Spanish/English headings (Aceptación/Límites/Supuestos), repair the collapsed nested bullets in "Casos borde" and "Supuestos" (which ends in a stray "- *."), and define or drop the [EXPLICIT]/[INFERENCE]/[SUPUESTO] tag vocabulary and undefined "ICM Layer 3" jargon.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~40-line body is lean with no concept explanations, but tokens are spent on per-line evidence tags ([EXPLICIT]/[INFERENCE]/[DOC]), undefined "ICM Layer 3" jargon, and a Spanish boilerplate "Packet" block that earn no operational value. Not a 5 because that clutter could be trimmed; not a 3 because there is no real padding or over-explanation. | 4 / 5 |
Actionability | The Procedure gives real file pointers ("open the matching references/ playbook") but the steps themselves are abstract — "Apply its decision tables; pick the strategy explicitly" and "tag every claim" give no commands, tag vocabulary, or entry-point invocation. The concrete commands (scaffold_spec.py/validate_spec.py usage) live only in the reference file, not in the body, leaving the body's own guidance incomplete. | 3 / 5 |
Workflow Clarity | A three-step sequence exists and step 3 says "Validate against the Quality Criteria", but the checkpoint is implicit — no command, no pass/fail criterion, and the available validator (validate_spec.py) or check.sh green/red loop is never surfaced in the body. This matches the anchor of steps listed with validation gaps rather than the explicit-checkpoint anchor. | 3 / 5 |
Progressive Disclosure | The two body references (references/full-playbook.md, references/full-skill-yaml-shape.md) are real, one level deep, and listed in a clean table — but the "Packet" block advertises knowledge/, prompts/, examples/, agents/, templates/ directories and a scripts/wire_packet.py generator that do not exist in the actual bundle, and the reference playbook itself points at further missing files (schemas/skill-spec.schema.json, templates/, examples/). Against the actual bundle structure, organization is only partially sound. | 3 / 5 |
Total | 13 / 20 Passed |