Content
78%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 strong, command-dense operational guide with an explicit lifecycle sequence and useful error-state guidance. Its two real gaps are the missing recovery loop beyond error detection, and complete failure to reference the two helper scripts shipped in the bundle, whose functionality is duplicated inline.
Suggestions
Reference the bundled helpers where their logic is inlined, e.g. 'For structured output use scripts/poll-status.sh <workdir> (enriches status with budget and council data)' and 'Format channel updates with scripts/format-progress.sh', instead of duplicating that logic in prose.
Add an explicit recovery loop for error states: what to do when 'failed tasks > 0' or status is 'unknown' (e.g. run 'loki logs --tail 50', inspect the failing task, then pause/resume or restart with justification).
Trim low-value JSON field documentation (e.g. 'version', 'dashboard_url', 'pid') to the fields actually needed for reporting decisions (status, phase, iteration, task_counts, elapsed_time).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is almost entirely commands, flags, and field documentation with no concept explanations Claude already knows — fitting 'efficient; minor instances of over-explanation that could be trimmed'. It is not a 5 because the 9-field JSON schema listing includes fields Claude rarely needs (e.g. 'version', 'dashboard_url') and 'Report final task counts, elapsed time, and duration' is partially redundant. | 4 / 5 |
Actionability | Every section gives copy-paste-ready concrete commands: 'loki start <prd-path> --bg --yes --no-dashboard', the poll command with workdir, all five control commands, and numbered completion steps including 'git log --oneline -20' and 'cat .loki/council/report.md'. This matches 'fully executable; copy-paste ready code or commands; specific examples cover the common cases' — start, monitor, report, control, and completion are all covered. | 5 / 5 |
Workflow Clarity | The full lifecycle is clearly sequenced (start with mandatory flags → poll every 30s → report transitions/counts/errors → control commands → numbered completion checklist), and error-recovery guidance exists ('If session crashes, check loki logs before restarting', flagging 'failed tasks > 0, status is unknown'). It falls short of 5 because the error handling stops at detection: there is no explicit recover/fix/retry loop for failed tasks or unknown status, only surfacing them to the channel. | 4 / 5 |
Progressive Disclosure | The body itself is well-organized with clear section headers, but the bundle provides scripts/poll-status.sh and scripts/format-progress.sh that automate exactly the monitoring/reporting logic the body inlines, and neither script is ever referenced — the bundle structure is unusable to Claude. This fits 'some structure but could be better organized' with content that should live in (or at least defer to) separate files left inline; it is above 2 because the body's own structure is solid and nothing is buried or nested. | 3 / 5 |
Total | 16 / 20 Passed |