Agent skill
pr-triage
Context-efficient PR comment triage. Evaluate, decide, act. Fix important issues, resolve the rest silently.
Install this agent skill to your Project
npx add-skill https://github.com/joelhooks/swarm-tools/tree/main/.opencode/skill/pr-triage
SKILL.md
PR Comment Triage - Evaluate → Decide → Act
Philosophy
Replies are SECONDARY to addressing concerns.
- Important issue? FIX IT → reply with commit ref → resolve
- Not important? RESOLVE SILENTLY → no reply needed
- Don't reply to every comment - that's noise
The Workflow
┌─────────────────────────────────────────────┐
│ EVALUATE → DECIDE → ACT │
├─────────────────────────────────────────────┤
│ │
│ 1. FETCH UNREPLIED (metadata only) │
│ → Get root comments without replies │
│ → ~100 bytes/comment, paginated │
│ │
│ 2. EVALUATE each comment │
│ → Fetch body only if path looks important│
│ → Skip: metadata files, style nits │
│ → Check: security, correctness, tests │
│ │
│ 3. DECIDE action │
│ → FIX: implement change, reply, resolve │
│ → RESOLVE: close silently, no reply │
│ → DEFER: create cell, resolve │
│ │
│ 4. ACT │
│ → Fix issues in code │
│ → Resolve threads (not reply) │
│ → Reply ONLY when you fixed something │
│ │
└─────────────────────────────────────────────┘
Decision Matrix
| Comment Type | Action | Reply? |
|---|---|---|
| Security/correctness bug | FIX → reply with commit | ✅ Yes |
| Valid improvement, in scope | FIX → reply with commit | ✅ Yes |
| Valid but out of scope | Create cell → resolve | ❌ No |
| Style/formatting nit | Resolve silently | ❌ No |
| Metadata file (.jsonl, etc) | Resolve silently | ❌ No |
| Already fixed | Reply with commit → resolve | ✅ Yes |
| Disagree with suggestion | Resolve silently | ❌ No |
SDK Commands
# Get unreplied root comments (start here)
bun run scripts/pr-comments.ts unreplied owner/repo 42
# Evaluate: fetch body for specific comment
bun run scripts/pr-comments.ts expand owner/repo 123456
# Act: resolve without reply (preferred)
bun run scripts/pr-comments.ts resolve owner/repo 42 123456
# Act: reply then resolve (only when you fixed something)
bun run scripts/pr-comments.ts reply owner/repo 42 123456 "✅ Fixed in abc123"
# Helpers
bun run scripts/pr-comments.ts summary owner/repo 42 # File-level overview
bun run scripts/pr-comments.ts list owner/repo 42 # All metadata
Quick Triage Pattern
import { fetchMetadata, fetchBody, resolveThread, reply, getThreadId } from "./scripts/pr-comments.ts";
const comments = await fetchMetadata("owner/repo", 42);
// Find unreplied root comments
const repliedTo = new Set(comments.filter(c => c.inReplyToId).map(c => c.inReplyToId));
const unreplied = comments.filter(c => !c.inReplyToId && !repliedTo.has(c.id));
for (const c of unreplied) {
// Skip metadata files - resolve silently
if (c.path.endsWith('.jsonl') || c.path.includes('.hive/')) {
const threadId = await getThreadId("owner/repo", 42, c.id);
if (threadId) await resolveThread("owner/repo", threadId);
continue;
}
// Evaluate important files
const full = await fetchBody("owner/repo", c.id);
if (full.body.includes('Critical') || full.body.includes('security')) {
// FIX IT, then reply
// ... implement fix ...
await reply("owner/repo", 42, c.id, "✅ Fixed in abc123");
}
// Resolve either way
const threadId = await getThreadId("owner/repo", 42, c.id);
if (threadId) await resolveThread("owner/repo", threadId);
}
Skip These (Resolve Silently)
.hive/issues.jsonl- auto-generated metadata.hive/memories.jsonl- auto-generated metadata- Changeset formatting suggestions
- Import ordering nits
- "Add tracking issue" for intentional skips
- Style preferences you disagree with
Fix These (Reply + Resolve)
- Security vulnerabilities
- Correctness bugs
- Missing error handling
- Test coverage gaps (if valid)
- Type safety issues
Context Budget
| Action | Context Cost |
|---|---|
unreplied |
~100 bytes/comment |
expand (1 comment) |
~5KB |
resolve |
0 (GraphQL mutation) |
reply |
~200 bytes |
Rule: Fetch <10 bodies per triage session.
References
scripts/pr-comments.ts- Full SDK with Zod schemasreferences/gh-api-patterns.md- Raw jq patterns, GraphQL, pagination
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
swarm-coordination
Multi-agent coordination patterns for OpenCode swarm workflows. Use when work benefits from parallelization or coordination. Covers: decomposition, worker spawning, file reservations, progress tracking, and review loops.
swarm-cli
Swarm CLI commands for workers - hivemind memory, hive tasks, swarmmail coordination. Use when working in a swarm context. Covers: swarm memory (find/store/get/stats), swarm cells (query/create/update/close), and coordination commands.
ralph-supervisor
Ralph loop pattern - Claude supervises while Codex (gpt-5.3-codex) executes implementation work. Use for autonomous coding loops with fresh context per iteration, validation gates, and git-backed persistence. Tools: ralph_init, ralph_story, ralph_iterate, ralph_loop, ralph_status, ralph_cancel, ralph_review.
always-on-guidance
Always-on rule-oriented guidance for claude-plugin agents. Use to align behavior, tool usage, and model-specific defaults while avoiding deprecated bd/cass references. Related skills: swarm-coordination, testing-patterns.
swarm-coordination
Multi-agent coordination patterns for OpenCode swarm workflows. Use when working on complex tasks that benefit from parallelization, when coordinating multiple agents, or when managing task decomposition. Do NOT use for simple single-agent tasks.
hive-workflow
Issue tracking and task management using the hive system. Use when creating, updating, or managing work items. Use when you need to track bugs, features, tasks, or epics. Do NOT use for simple one-off questions or explorations.
Didn't find tool you were looking for?