AgentSkillsCN

inspect

深入探究特定故事的状态——包括阶段、失败原因、依赖项、上一次代理输出、需求验证，以及建议的下一步行动

SKILL.md
--- frontmatter
name: inspect
description: Deep-dive into a specific story's state — stage, failures, dependencies, last agent output, requirement verification, and suggested next action
argument-hint: "<US-XXX>"
allowed-tools:
  - Read
  - Glob
  - Grep
  - Bash(jq *)

Story Inspector

Input

Story ID from $ARGUMENTS (format: US-XXX). If empty → print usage. If prd/$ARGUMENTS.json missing → list available stories.

Data Collection

  1. •Story JSON: prd/<ID>.json — all fields. .stage is top-level.
  2. •Tracker: prd/<ID>.md
  3. •Dependencies: For each in dependencies array → read JSON for id, title, .stage
  4. •Dependents: Search all prd/US-*.json for stories listing this ID in dependencies
  5. •Logs: Grep .log/run-*/run.md for story ID
  6. •Pipeline: .state/pipeline.json → check if currentStory matches

Output

code
=== Story: <ID> — <title> ===
Area: <area> | Priority: <priority> | Risk: <riskTier> | Size: <sizing> | Autonomy: <autonomy>

Stage: <stage> | Deploy attempts: <status.deploy.attempts>/3 | Deploy notes: <notes>
Pipeline: <"Active" / "Not active">

Requirements: | REQ ID | Implemented | Verified |
Dependencies (upstream): <ID> — <title> [<stage>] BLOCKING/OK
Dependents (downstream): <ID> — <title> [<stage>] WAITING/OK
Deploy Safety: strategy, healthChecks, rollbackTrigger, rollbackCommand, verification
Recent Logs (max 5): Run <ID>: <agent> | <outcome>
Tracker Notes: <summary from .md>

Next Action

  • •build/review → ./orchestrator.sh run --tool claude --story <ID>
  • •deploy (gated) → ./orchestrator.sh run --tool claude --story <ID> --approve-deploy <ID>
  • •complete → No action
  • •blocked → Review notes, /reset-story <ID>
  • •Upstream incomplete → "Blocked by <US-XXX> at <stage>"