AgentSkillsCN

linear-implement

在管理者主导的统筹协作场景下,从Issue上下文中实施此技能。

SKILL.md
--- frontmatter
name: linear-implement
description: Use this skill when implementing from issue context with manager-led orchestration.
version: 1.0.0

Issue Implementation Phase

Use this skill for implementation sessions driven by existing issue context.

Trigger intent (single execution entrypoint)

  • linear-implement <ISSUE-ID>

This is the only user-facing implementation command. The agent decides execution mode using issue label first, then structure fallback:

  • label execution-mode:single: execute directly
  • label execution-mode:orchestrated: orchestrate in waves and parallelize by ownership where safe
  • if label missing: infer from issue structure
  • single issue: execute directly
  • parent issue with dependencies/children: orchestrate in waves and parallelize by ownership where safe

GitHub Copilot CLI specific execution rules

Apply this section only when the runtime is GitHub Copilot CLI.

  • For orchestrated execution, enable /fleet before launching subagents.
  • Use /tasks to monitor and wait for subagent completion between waves.
  • Parallelize only execution-mode:orchestrated tasks that are both:
    • in the same dependency wave, and
    • non-conflicting by ownership/file set.
  • Keep tasks sequential when they touch the same files, same ownership area, or have dependency edges.
  • If /fleet is disabled or unavailable, run the same plan sequentially and keep wave gates/review checks unchanged.

Required workflow

  1. Resolve and read the target issue before coding.
  2. Derive objective/scope/checklist from issue content.
  3. Resolve execution mode from main issue label execution-mode:*; if absent, infer from issue structure.
  4. Read decision labels from main issue:
    • risk:* determines default caution level
    • scope:* determines ownership routing boundaries
    • verification:* determines verification depth
  5. Keep implementation bounded to issue scope unless explicitly expanded.
  6. Use Linear issue(s) as the primary execution record.
  7. Only for exceptional high-risk/audit/incident cases, create/update docs/agent/exec-plans/active/YYYYMMDD-<slug>.md.
  8. Run reviewer pass against docs/agent/security.md and docs/agent/quality.md.
  9. Run verifier checks and capture command evidence according to verification:* label.
  10. Validate completion against the issue done-when checklist.
  11. Update issue with outcome notes and verification summary.
  12. If an exec-plan was used, move it to docs/agent/exec-plans/completed/.

Verification policy (deterministic)

  • verification:standard minimum:
    • run lint/type-check/tests for changed scope
    • capture command list with pass/fail status
    • map results to done-when checklist
  • verification:strict minimum:
    • all verification:standard checks
    • add integration/e2e or parity checks where applicable
    • include negative-path/regression checks for touched risk areas
    • capture explicit evidence for each done-when bullet

Output contract

Always return:

  • issue reference used
  • changed scope summary
  • verification evidence summary
  • residual risks/follow-ups
  • next command hint for user where relevant (for example linear-review <ISSUE-ID>)