feat(agents): add Claude Code routines scaffolding#834
Merged
Conversation
Add the committed half of the issue-triage routines setup: triage prompt and cloud environment setup script. Also add a GitHub Actions bridge that fires the routine's `/fire` endpoint on `issues.opened` / `reopened` so response happens in minutes rather than at the next scheduled run. See `.agents/routines/README.md` for the identity and setup-order checklist. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add pre-classification (RFC/epic/tracking → no auto-PR), scope-bucket labeling (gh label list, map to existing labels, repo-specific buckets), and milestone-assignment (gh api milestones, suggest fit, apply to auto-PRs). Updates the structured triage comment with **Bucket(s):** and **Suggested milestone:** lines. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Address review feedback from security / prompt-engineering / code review / DX / ad-tech product agents: - Bridge workflow wraps untrusted issue body in a fenced "data, not instructions" block; 8KB truncation; NUL-byte strip; curl exit handled separately from HTTP status; redacts bearer-looking substrings; tighter bot filter; concurrency group; 2-min timeout; set -euo pipefail. - environment-setup.sh uses --ignore-scripts (npm) / upgrades pip / removes || true fallbacks / pins golangci-lint installer URL. - CODEOWNERS added so agent-authored PRs touching .agents/, .github/, deps, release tooling, and (adcp-go) TEE-bound paths require human review before merge. - Triage prompt: RFC/epic/tracking/child-of-open-parent pre-class (never auto-PR); author-association gate (drive-bys get comment- only); bucket labels must be a clear match (never invent); milestone only when target version named/linked/labeled; 1500-char comment cap; duplicate + open-PR checks; API-failure retry path; first-time-contributor "Thanks for filing!" framing. - adcp-go: added claude-bot-path-guard.yml that fails bot PRs touching identity/, router/pinhole*, router/metrics*, internal/sanitize/, go.mod/go.sum, .agents/, .github/workflows/, release-please-*. Security-sensitive issues never get PRs and never describe the vector in the public comment. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
First live run on adcp surfaced that the routine was posting boilerplate triage comments on well-structured RFCs/features from maintainers — pure noise. Add a silent path: apply claude-triaged + bucket labels with no comment when the comment would add no signal (classification is feature/RFC/epic, author is established, body is well-structured, issue already has on-target label). Still comment on drive-bys, bugs, needs-info, and anything that surfaces a duplicate / related PR / cross-repo redirect. For adcp-go specifically: security-sensitive bugs and performance issues are never silent-triaged — the withheld-vector comment *is* the signal there. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Port the v2 triage pattern from adcp to this repo: - Scaffolds 6 expert subagents under .claude/agents/ tailored to this repo's scope: ad-tech-protocol, adtech-product, code-reviewer, dx-expert, docs-expert, security-reviewer. Skipped experts (prompt-engineer, education, etc.) that don't apply to a library / SDK context. - Rewrites triage-prompt.md to match adcp v2: bucket-driven expert panel, four outcomes (clarify / flag / execute-PR / defer), concurrency check (10-min comment-presence lock), coverage checklist at synthesis step. - Drops silent-triage default and NONE-author PR gate — drive-by bugs welcome when small and correct; CODEOWNERS still gates merge. - Repo-specific adaptations: bucket list, expert panel mapping, test commands, forbidden paths. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Mirror of adcontextprotocol/adcp#2957: 1. Already-engaged check before expert consultation. Silent-defer when the issue is assigned to a repo member, has an open PR referencing it, or has a repo-member comment in the last 7 days — a human is probably already working on it in a Conductor workspace / drafts / Slack and the bot's synthesis would pre-frame their thinking. 2. Tighten "never create labels" rule: must run gh label list first, apply only names returned, never POST to /labels, default not-apply on uncertainty. Missing bucket labels go in the comment body plus the run summary. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Mirror of adcontextprotocol/adcp#2958. Flip the default from "flag unless obviously a small bug" to "execute unless breaking or ambiguous." - Drop <150 line scope cap - Drop classification-only-Bug-or-Doc gate - Replace with single non-breaking-vs-breaking binary - "When in doubt: Execute" mantra (inverted on adcp-go — TEE posture means when in doubt, Flag) CODEOWNERS + human review still gate every merge. Generated files, pinned deps, TEE-bound paths (adcp-go) still always Flag. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
.agents/routines/triage-prompt.md(behavior),environment-setup.sh(cloud env install), andREADME.md(identity + setup checklist)..github/workflows/claude-issue-triage.yml— a minimal bridge that POSTs new issues to a per-routine/fireendpoint so the triage routine reacts within minutes instead of waiting for its next scheduled run.What this does NOT do
/scheduleskill) to set prompt/repo/environment/schedule/API trigger. The committed prompt is what the routine points at.CLAUDE_ROUTINE_TRIAGE_URLandCLAUDE_ROUTINE_TRIAGE_TOKENrepo secrets are set (logs a warning and exits 0).Test plan
.github/workflows/claude-issue-triage.yml.ci,changeset-check, etc.) still passes.🤖 Generated with Claude Code