Greptile Summary

This PR updates skills/coding-agent/SKILL.md to replace the worker-owned openclaw message send completion pattern with a parent-owned model: workers print a structured CODING_AGENT_RESULT_START/END block to stdout, and the parent assistant reads it via process action:log to compose the user-facing report.

Confidence Score: 4/5

Documentation-only change; safe to merge with the minor template inconsistency addressed.

Only P2 findings — a minor inconsistency in placeholder quoting between the canonical CODING_AGENT_RESULT block template and the embedded Quick Start example. No logic, security, or correctness issues.

skills/coding-agent/SKILL.md — template quoting inconsistency in the CODING_AGENT_RESULT block.

This is a comment left during a code review.
Path: skills/coding-agent/SKILL.md
Line: 116-121

Comment:
**Template inconsistency between fenced block and embedded Quick Start**

The canonical template (lines 116–121) uses quoted placeholder values — `"not run: <reason>"` and `"none"` — but the embedded version in the Quick Start bash command (lines 143–148) drops the quotes: `not run: reason` / `none`. Workers that follow one variant will produce output in a different format from workers following the other, which could complicate log parsing with `CODING_AGENT_RESULT_START/END`.

Consider normalizing both locations to the simpler unquoted style used in the Quick Start.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "docs(skills): make coding-agent completi..." | Re-trigger Greptile}

Codex review: needs maintainer review before merge.

What this changes:

This PR rewrites skills/coding-agent/SKILL.md so background coding-agent workers print a structured stdout result block while the parent assistant captures the sessionId, inspects process logs/status, and reports completion or failure itself.

Maintainer follow-up before merge:

Review this as a focused maintainer docs/product-contract PR: either accept the parent-owned completion model in skills/coding-agent/SKILL.md with the stdout result block and no routine routing details in worker prompts, or keep the current direct-send model and close the PR with that explicit policy decision. If accepted, docs-only validation such as git diff --check is the appropriate floor.

Best possible solution:

Review this as a focused maintainer docs/product-contract PR: either accept the parent-owned completion model in skills/coding-agent/SKILL.md with the stdout result block and no routine routing details in worker prompts, or keep the current direct-send model and close the PR with that explicit policy decision. If accepted, docs-only validation such as git diff --check is the appropriate floor.

What I checked:

• Current skill requires worker self-notification: Current main says the completion notification must use openclaw message send, not system event/heartbeat, and tells the parent to require the worker to notify the user directly when it finishes. (skills/coding-agent/SKILL.md:3, d30ba5351ed1)
• Mandatory pattern conflicts with the PR intent: The current mandatory pattern captures channel/account/reply routing, injects it into the worker prompt, and requires the worker to call openclaw message send on completion; there is no CODING_AGENT_RESULT stdout block on main. (skills/coding-agent/SKILL.md:91, d30ba5351ed1)
• PR diff is focused on the docs contract: The submitted diff only changes skills/coding-agent/SKILL.md, replacing notification-route injection with sessionId capture, process log/poll inspection, and a CODING_AGENT_RESULT_START / CODING_AGENT_RESULT_END final stdout block. (skills/coding-agent/SKILL.md:1, e73c399b87a3)
• Relevant docs support reviewing the completion model carefully: Background task docs describe terminal task notification as push-driven, with direct delivery or session-queued delivery plus immediate heartbeat wake; that supports the PR's premise but does not mean the coding-agent skill on main already implements the proposed parent-owned flow. Public docs: docs/automation/tasks.md. (docs/automation/tasks.md:162, d30ba5351ed1)
• Security and supply-chain pass: The PR changes one bundled skill markdown file and does not touch workflows, package metadata, lockfiles, external action refs, installer scripts, dependencies, or generated/vendor code. The security-relevant effect is prompt-routing guidance: it removes routine channel/account/reply IDs from worker prompts unless explicitly needed. (skills/coding-agent/SKILL.md:1, e73c399b87a3)
• Feature history: git log --follow and git blame show the current bundled coding-agent skill, including the direct openclaw message send completion requirement, was added on main by commit e1fd27f. (skills/coding-agent/SKILL.md:1, e1fd27fb24ae)

Likely related people:

• steipete: git log --follow and line blame show commit e1fd27f added skills/coding-agent/SKILL.md and the current direct openclaw message send completion guidance. (role: introduced behavior / current-main owner; confidence: high; commits: e1fd27fb24ae; files: skills/coding-agent/SKILL.md)

Remaining risk / open question:

• The PR is docs-only, but it changes the operational contract for a bundled skill. A maintainer should confirm parent-owned completion is the desired policy for low-level background bash workers and that the automatic wake/watchdog wording matches runtime behavior.

Codex review notes: model gpt-5.5, reasoning high; reviewed against d30ba5351ed1.

Thanks for the skill-docs update. Closing this as low-signal process-doc churn without a maintainer-owned workflow change; please resubmit with a concrete current workflow gap if this still needs to change.