skill(gh-issues): split oversized references from SKILL.md

[ShionEria]

Summary

• keep the gh-issues skill workflow in the main SKILL.md
• move bulky reference-heavy prompt and review details into references/
• reduce the main SKILL.md below the 500-line best-practice limit

Testing

• wc -l skills/gh-issues/SKILL.md -> 483

Closes #50972

[openclaw-barnacle]

Closing this PR because the author has more than 10 active PRs in this repo. Please reduce the active PR queue and reopen or resubmit once it is back under the limit. You can close your own PRs to get back under the limit.

[greptile-apps]

Greptile Summary

This PR refactors the gh-issues skill by extracting three large inline blocks from SKILL.md into dedicated reference files under skills/gh-issues/references/, bringing the main file from ~550+ lines down to 483 — within the 500-line best-practice limit. The existing workflow logic is unchanged; SKILL.md now delegates to the reference files with "Read that file when building the task" instructions.

• references/fix-subagent-prompt.md — Phase 5 fix sub-agent prompt template, faithfully migrated
• references/review-handler.md — Phase 6 review-source fetch rules, actionability criteria, and review-fix sub-agent prompt, faithfully migrated
• references/reporting-and-watch.md — results summary tables, watch-mode state retention rules, and the orchestrator-level Telegram notification template, mostly faithful but the tool invocation details for the orchestrator's final-summary notification (action: send, channel: telegram, target: {notify-channel}) were not carried over — only the message body format was preserved. The sub-agent's per-PR notification in fix-subagent-prompt.md still includes the full invocation details, so single-issue notifications are unaffected.

Confidence Score: 4/5

• Safe to merge with minor follow-up: the orchestrator-level Telegram notification is missing tool invocation details in the new reference file.
• The refactor is a straightforward, faithful content migration. All workflow logic is preserved. The one functional gap — omitted message tool parameters in reporting-and-watch.md — only affects the orchestrator's final summary notification, not core fix-agent or review-agent behavior, and LLM agents may infer correct usage from nearby context in fix-subagent-prompt.md.
• skills/gh-issues/references/reporting-and-watch.md — orchestrator Telegram notification section is missing the explicit message tool invocation details.

Comments Outside Diff (1)

1. skills/gh-issues/references/reporting-and-watch.md, line 668-682 (link)

   Orchestrator notification tool invocation details dropped

   The original SKILL.md included explicit instructions for the orchestrator to use the message tool with specific parameters when sending the final summary notification. The new reporting-and-watch.md preserves the message format but omits the tool invocation details (action: "send", channel: "telegram", target: "{notify-channel}").

   The sub-agent in fix-subagent-prompt.md (step 10) still has these explicit instructions, but the orchestrator-level final-summary notification path no longer has them. The relevant section currently reads:

   If `--notify-channel` is set, send the final summary with only successfully opened PRs:
   

   Without specifying how to send it (tool name, action, channel, target parameters). Consider restoring the tool invocation details, e.g.:

   text
   ✅ GitHub Issues Processed

   Processed {N} issues: {success} PRs opened, {failed} failed, {skipped} skipped.

   {PR_LIST}

   
   Where `PR_LIST` uses:
   
   ```text
   • #{issue_number}: {PR_url} ({notes})
   

   
   <details><summary>Prompt To Fix With AI</summary>
   
   `````markdown
   This is a comment left during a code review.
   Path: skills/gh-issues/references/reporting-and-watch.md
   Line: 668-682
   
   Comment:
   **Orchestrator notification tool invocation details dropped**
   
   The original `SKILL.md` included explicit instructions for the orchestrator to use the `message` tool with specific parameters when sending the final summary notification. The new `reporting-and-watch.md` preserves the message format but omits the tool invocation details (`action: "send"`, `channel: "telegram"`, `target: "{notify-channel}"`).
   
   The sub-agent in `fix-subagent-prompt.md` (step 10) still has these explicit instructions, but the **orchestrator-level** final-summary notification path no longer has them. The relevant section currently reads:
   
   ```text
   If `--notify-channel` is set, send the final summary with only successfully opened PRs:
   

   Without specifying how to send it (tool name, action, channel, target parameters). Consider restoring the tool invocation details, e.g.:

   text
   ✅ GitHub Issues Processed

   Processed {N} issues: {success} PRs opened, {failed} failed, {skipped} skipped.

   {PR_LIST}

   
   Where `PR_LIST` uses:
   
   ```text
   • #{issue_number}: {PR_url} ({notes})
   

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

Prompt To Fix All With AI

This is a comment left during a code review.
Path: skills/gh-issues/references/reporting-and-watch.md
Line: 668-682

Comment:
**Orchestrator notification tool invocation details dropped**

The original `SKILL.md` included explicit instructions for the orchestrator to use the `message` tool with specific parameters when sending the final summary notification. The new `reporting-and-watch.md` preserves the message format but omits the tool invocation details (`action: "send"`, `channel: "telegram"`, `target: "{notify-channel}"`).

The sub-agent in `fix-subagent-prompt.md` (step 10) still has these explicit instructions, but the **orchestrator-level** final-summary notification path no longer has them. The relevant section currently reads:

```text
If `--notify-channel` is set, send the final summary with only successfully opened PRs:
```

Without specifying *how* to send it (tool name, action, channel, target parameters). Consider restoring the tool invocation details, e.g.:

```suggestion
If `--notify-channel` is set, send the final summary with only successfully opened PRs using the `message` tool:

- action: `send`
- channel: `telegram`
- target: `{notify-channel}`
- message:

```text
✅ GitHub Issues Processed

Processed {N} issues: {success} PRs opened, {failed} failed, {skipped} skipped.

{PR_LIST}
```

Where `PR_LIST` uses:

```text
• #{issue_number}: {PR_url} ({notes})
```
```

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

_{Last reviewed commit: "skill(gh-issues): sp..."}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ef3d144c97

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Restore the message-tool invocation for summary notifications

This extracted section only preserves the notification body template. skills/gh-issues/SKILL.md:371-372 now tells Phase 5 to rely on this file, but unlike the previous inline block it no longer says to call the message tool or which channel/target to send to. In runs that set --notify-channel, the orchestrator now has no concrete send instructions, so it's likely to just print the summary locally instead of delivering the Telegram notification.

Useful? React with 👍 / 👎.