Index of every page on this site: llms.txt
All skills
harlan-zw avatar

/pr-triage

@2a530d0
by Harlan Wiltonharlan-zw/harlan-agent-kit14 stars
1
Run it yourself

Triage all open pull requests in Harlan-owned repositories. Use to repair, rank, sign off, or decide merge order across the PR backlog.

  • 3 files
  • 5.5 KB
  • MIT
  • Updated last month
  • GitHub
Use this Skill: https://skilld.dev/gh/harlan-zw/harlan-agent-kit/pr-triage

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈37 tokens for metadata: the name and description. ≈796 when used: this file. ≈51 more on demand in 1 file.

Description uses 1.8% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈147 characters, or 1.8%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

PR Triage

Turn the owned PR backlog into a ranked merge queue. Never merge.

harlan-github-agent merges pull requests labelled harlan-agent-auto-merge on its own. This skill never does. See auto merge.

Worktree isolation

Discovery stays read only and may use the primary checkout. Before Repair, read and follow the worktree isolation contract. Keep mutation in a task-owned wt worktree with a live claim. Keep the primary checkout read only.

Load contracts

Read these completely before discovery:

  1. ../adversarial-review/SKILL.md, the complete workflow for one PR.

The adversarial-review Skill loads personal pr, Brundlefly im-part-human, and unit-tests when required. Do not duplicate those rules here.

Discover

Run scripts/discover-prs.sh from this skill directory.

It returns every human-authored open PR in the allowed base repositories. Exclude GitHub Apps and bot accounts unless the user requests --include-bots.

Skip automated authors before dispatch, review attempts, or comments. Do not post a skip status.

For an author outside configured writable_pr_authors, wait for exact-Revision Review Approval in harlan-github-agent. Do not create a status comment before Approval. Require separate fix Approval after open findings and before edits.

If discovery reaches GitHub's 1,000 result cap, stop before mutations and report incomplete discovery.

Process the backlog

Process one PR at a time. Parallelize read-only GitHub queries when useful.

For each PR, run the complete adversarial-review workflow. A PR is unfinished until its marked bot comment is confirmed on GitHub.

Treat one trusted marked comment for the current head commit as the active review. Never dispatch a second agent for that head commit.

For a personal repository mapped under ~/sites, use take-ownership by convention unless repository policy or the user disables it. Other PRs stop after adversarial review.

Continue to the next PR after a confirmed READY, PENDING, or BLOCKED status. Preserve exact failure evidence.

Finish repairs across the backlog, then revisit PENDING pull requests once. Never sign off a stale SHA.

When required CI also fails on the current base of an owned repository, dispatch one focused baseline repair worker. Open its repair pull request, link it from the original PR status, and revisit the original after merge.

Return the merge queue

Predict attention for this handoff. Lead with the consequential merge decision or blocker, with its required dependencies. Keep routine proof brief and link detailed evidence. Preserve the complete queue and every safety reason.

List READY pull requests first, sorted by confidence. Then list PENDING and BLOCKED pull requests without confidence scores.

Confidence | PR | Outcome | Checked | Next action
96/100 high | owner/repo#42 | READY | Base, review, tests, CI | Human merge decision
— | owner/repo#17 | BLOCKED | Review complete; CI failed | Fix Linux job

Include skipped PRs and the exact safety reason. Never merge unless the user separately asks.

Source: SKILL.md on GitHub

Skill behaviors

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

Third-party checks

No third-party reports yet.

Provenance

Signed by skilld at 2a530d0. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 days ago.

Activeupdated last month

README badge

README badge for harlan-zw/harlan-agent-kit/pr-triage