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 dense but genuinely actionable procedure: concrete commands, clearly sequenced modes with repeat-until-settled loops and refusal handling, and a clearly signaled one-level-deep reference to the governing law file. Its main cost is prose density — very long conditional sentences, heavy cross-reference boilerplate, and row formats deferred to the external file make it harder to follow than its content warrants.
Suggestions
Rewrite Sweep step 3's nested-conditional run-on sentences (duplicates, missing paths, supersession) as short decision-table rows or bulleted condition→row pairs so each rule can be scanned in one pass.
Replace the ~30 scattered "as chips.md's X says" citations with one compact map of law sections near the top (e.g., Sweep, Offer, Priority, Milestone, Cards, dup bullet), referencing it from the steps.
Show the exact ledger row syntax for the recurring outcomes (updated, dismissed done/superseded/dup/stopped/declined, folded, carded, landed) as a small reference table or example block instead of describing their tokens in prose.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence and explains no background concepts, but it could be tightened substantially: Sweep step 3 packs nested conditional branches into single run-on sentences (e.g., the 100+ word duplicate-handling bullet), and the phrase "as chips.md's ... says" is repeated roughly thirty times. This matches 'mostly efficient but includes some unnecessary explanation or could be tightened' rather than 4, because the density itself forces re-reading and the repeated citation boilerplate is trimmable. | 3 / 5 |
Actionability | Concrete, runnable commands appear throughout — "node .agents/pstack/decisions-check.mjs <command>", "chips cleanup --sessions <file> --session <id>", "append <ledger> chip \"<id> <title>\" \"<finding>\" \"<tokens>\" <result>", "chips brief <id>", "git ls-files" — matching 'mostly executable guidance with minor gaps'. It is not 5 because the exact row/token syntax for outcomes like 'updated' and 'dismissed declined' is deferred to the external chips.md rather than shown, leaving the reader to assemble rows from prose. | 4 / 5 |
Workflow Clarity | Each mode is a numbered sequence with explicit feedback loops and validation: "Repeat step 2 for what step 3 changed", re-running the board read "until the Do list shows only the card pass or nothing you can apply", and "When append refuses a batch, fix or drop the refused row as its message says and append the rest". This matches 'clear sequence with most checkpoints present; minor validation gaps' — it is not 5 because checkpoints like step 6's error handling are buried inside single dense sentences rather than standing out as explicit validate/fix/retry steps. | 4 / 5 |
Progressive Disclosure | No bundle files exist, and the single external reference is clearly signaled one level deep: "The law is `.agents/pstack/rules/chips.md` ... Read it in full before this session's first ledger write", with an up-front mode table and per-mode sections giving good navigation. This matches 'good structure; most content is appropriately placed; references mostly clear; minor organization gaps' — not 5 because the long Sweep section mixes orchestration with rule detail that the referenced law file also holds, and internal anchors exist only for mode names, not for the sweep's sub-passes. | 4 / 5 |
Total | 15 / 20 Passed |