Content
75%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 delivers concrete, well-sequenced MCP-driven guidance with decision rubrics and a useful troubleshooting section, scoring solidly across all dimensions. The main uplift opportunities are moving the postmortem template and troubleshooting detail into a one-level reference file, trimming a few obvious-culture sentences, and specifying concrete tools/steps for the remaining vague checks (user impact estimation, rollback verification).
Suggestions
Move the 6-section postmortem structure and the three Troubleshooting deep-dives into a references/ file (e.g. POSTMORTEM-TEMPLATE.md, TROUBLESHOOTING.md) and reference them from the body, keeping SKILL.md as a lean overview.
Replace vague checks with concrete tool calls: for "Check if a rollback was performed and whether it resolved the issue" and "Estimate affected users based on traffic volume", specify which harness_list/harness_status queries or data sources answer them.
Trim sentences that restate knowledge Claude already has, e.g. "blameless culture requires acknowledging good responses" and "while the incident is fresh in the team's memory", and consolidate the duplicated harness_list deployment-pull blocks in Steps 3 and 5.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is largely operational (tool call templates, correlation heuristics, severity rubrics) with only minor over-explanation that could be trimmed, e.g. "blameless culture requires acknowledging good responses" and "Postmortems should be generated within 48 hours while the incident is fresh in the team's memory" — Claude already knows blameless postmortem conventions. This fits the 4 anchor ('minor instances of over-explanation') rather than 3, since the padding is a few sentences in a 157-line body, and not 5 because those sentences and the repeated 'Gather from the user' / duplicate harness_list blocks don't all earn their place. | 4 / 5 |
Actionability | Concrete MCP call templates with explicit parameters (harness_list with resource_type/org_id/project_id/status; harness_status), quantitative heuristics (2-hour correlation window, 48-hour postmortem window), and defined decision outputs (HIGH/MEDIUM/LOW confidence criteria, Critical/Major/Minor severity criteria, rollback/failover/circuit-breaker recommendations). It stops short of the 5 anchor because some guidance remains directive without a concrete method, e.g. "Estimate affected users based on traffic volume" and "Check if a rollback was performed", with no specified tool or query for those checks. | 4 / 5 |
Workflow Clarity | Steps 1-5 are clearly sequenced with three named sub-workflows, and the Troubleshooting section supplies error-recovery paths for each workflow's failure modes (no deployment found, missing services, missing timeline events) — genuine feedback loops. Not 5 because mid-workflow validation is implicit rather than explicit: e.g. the timeline construction and blast-radius mapping have no checkpoint to verify data completeness before drawing conclusions, fitting the 4 anchor ('most checkpoints present; minor validation gaps'). | 4 / 5 |
Progressive Disclosure | The single-file skill is well-organized into Instructions, Examples, Performance Notes, and Troubleshooting with no nesting. It falls short of the 5 anchor because at 157 lines some self-contained material is inlined that would sit better one level deep — notably the 6-section postmortem template and the three troubleshooting deep-dives — matching the 4 anchor ('most content appropriately placed; minor organization gaps'). | 4 / 5 |
Total | 16 / 20 Passed |