cawplan-risk-assessment

CawPlan Risk Assessment — A5 Release Risk Assessment

cawplan skill check

Hard Gates (MUST)

1. Rules compute the level, AI writes the summary：risk_level Use compute response as authoritative; Agent must notrecompute on its own or silently downgrade CRITICAL。
2. Save requires double gate: show compute result → user confirmation → --body-file + --dry-run → submit gate → --confirm executes save。
3. Prohibitedto treat A3 execution progress,cawplan-plan-track schedule of risk/risk_reason as A5 conclusion;Prohibitedto claim this conclusion is equivalent to the Release Readiness of the A8 Ticket Report. Allowedto call versions track / tickets search / execution failures asread-only context enrichment, must be marked "not involved in rule computation".
4. Prohibited cawplan api, direct connection to TestRail, A1/A2/A4/A8 write operations.
5. User display: friendly field names;product_id/version_id/UUID hidden by default; Agent's own AskUserQuestion(not written into allowed-tools）。

Reference loading (MUST)

Timing	Read	Prohibited
Opening	None	One-time Read of all references
Parse context / compute / enrichment / interpret reasons	compute-rules.md	—
Output template / enrichment table / runs appendix	ux.md	together with save-rules.md Read
Save / override / handoff to A4·A7	save-rules.md	—

Allowed commands

Step	Command
Auxiliary	products list、versions list
Read-only enrichment (Step 4b)	versions track <product_id> <version_id>
Read-only enrichment (Step 4b)	tickets search --unique_ids <id1,id2,...>（Not required --time_range）
Read-only enrichment (Step 4b)	qa-insights testrail execution failures <product_id> <version_id> [--run-id] [--limit]
Threshold (optional)	qa-insights risk-rules get <product_id>
History	qa-insights risk-assessment get <product_id> <version_id>
Compute	qa-insights risk-assessment compute <product_id> <version_id> [--ticket-id] [--plan-mapping-ids] [--no-refresh-execution]
Save	qa-insights risk-assessment save <product_id> <version_id> --body-file <path> [--dry-run] --confirm

Workflow

Step	Action	Detail
0	Parse product / version; optional ticket_id	compute-rules.md §0
1	Intent triage: history only vs recompute vs save	compute-rules.md §Intent
2	Optional risk-rules get(user asks about threshold or HIGH reason)	compute-rules.md §Rules
3	Optional risk-assessment get(history only and recompute not requested)	compute-rules.md §Get
4	risk-assessment compute(refreshes execution data by default)	compute-rules.md §Compute
4b	Read-only enrichment (after compute or get; conditionally triggered by reason; can be parallel)	compute-rules.md §Enrich · ux.md §Enrich
5	Agent interpretation: summary + layered report;do not modify risk_level	ux.md §Output
6	Save decision gate (stop if user did not request save)	save-rules.md §Prompts
7	Assemble body-file → --dry-run → submit gate → --confirm	save-rules.md §Save
8	Save result + one handoff sentence at the end (A4/A7)	save-rules.md §Handoff

User explicitly says "only the last conclusion / no recompute" → skip Step 4, go Step 3 → 4b → 5。
User explicitly says "save" but no compute in this session → first Step 4 or confirm to reuse get record then save。

Output summary

Not saved: risk level + release/completeness context + coverage priority metrics + enrichment detail table + recommended actions + whether-to-save question.
Saved: append saved_at、saved_by、note、ai_summary。
Template see ux.md §Output、save-rules.md §Saved。

References

• compute-rules.md — context, compute/get, interpretation of level and reasons, enrichment triggering
• ux.md — Report template, enrichment table, runs appendix, recommended actions display
• save-rules.md — Save body, override, confirmation gate, A4/A7 connection