Conversation
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
git-subtree-dir: lib/slack-cli git-subtree-split: 0e5b91258f5d91aee4f2d17edbc3bab44bf5dc6b
Amp-Thread-ID: https://ampcode.com/threads/T-019d47cc-caf7-740e-a487-be1f0e3443f8 Co-authored-by: Amp <amp@ampcode.com>
…put drain cleanup Addresses Copilot review comments: - Extract sendSlackUpdate() so it can be called from interval AND finally block - Add 10s minimum throttle between LLM-synthesized Slack updates (avoids rate limits) - Call sendSlackUpdate() explicitly in finally block (was broken: interval cleared first) - Abort controller in finally to cancel inputDrainPromise reader Amp-Thread-ID: https://ampcode.com/threads/T-019d47cc-caf7-740e-a487-be1f0e3443f8 Co-authored-by: Amp <amp@ampcode.com>
… pooling Addresses Copilot review: client() was creating a new Client per request, defeating connection pooling. Amp-Thread-ID: https://ampcode.com/threads/T-019d47cc-caf7-740e-a487-be1f0e3443f8 Co-authored-by: Amp <amp@ampcode.com>
…nvalid markdown type Addresses Copilot review: Slack Block Kit does not support 'markdown' block type. Use 'section' block with 'mrkdwn' text object instead. Amp-Thread-ID: https://ampcode.com/threads/T-019d47cc-caf7-740e-a487-be1f0e3443f8 Co-authored-by: Amp <amp@ampcode.com>
…ersion, caching, and safe messaging Standalone utilities ported from lib/slack/ with zero project-specific deps: - parse-url: parseSlackUrl() + slackMessageUrlParse() with team extraction - ts-to-iso: slackTsToISO() timestamp conversion - parse-markdown: sync + async mrkdwn→markdown with optional resolver - safe-message: truncateSlackText() + truncateSlackBlocks() - cached-client: createCachedSlack() in-memory Map cache proxy Amp-Thread-ID: https://ampcode.com/threads/T-019d47cc-caf7-740e-a487-be1f0e3443f8 Co-authored-by: Amp <amp@ampcode.com>
…lack channel Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-019d56a9-781f-705a-9c1f-408916e8d206
Amp-Thread-ID: https://ampcode.com/threads/T-019d56a9-781f-705a-9c1f-408916e8d206 Co-authored-by: Amp <amp@ampcode.com>
- Add bot/task-user.ts: create/manage task-* system users with comfy-tasks group - Add bot/spawn-as-user.ts: custom spawnClaudeCodeProcess using sudo -u per task - Update bot/slack-bot.ts: integrate task user creation before agent spawn, add hourly cleanup, fix --continue IIFE bug (async fn was declared but never called) - Update lib/zChat.ts: replace require() with async import() for ESM compatibility, add initZChat() Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Replace SocketModeClient with Bun HTTP /slack/events endpoint - Verify requests with HMAC-SHA256 using SLACK_SIGNING_SECRET - Handle url_verification challenge for Slack App setup - Extract handleSlackEvent() for clean event dispatch - Caddy: add prbot.stukivx.xyz → :3475 route Webhook URL: https://prbot.stukivx.xyz/slack/events Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- webhook handler: dedup duplicate events by event_id (1h TTL) so Slack retries (x-slack-retry-num) no longer trigger 3x task execution - bot/index.ts: load .env.local with override semantics + log secret prefixes at startup to prevent stale shell env shadowing the file (regression from the 2026-04-24 signing-secret mismatch incident) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- fast-ack: post a "👀 受け取りました" placeholder before the intent LLM call so users see activity within ~1s instead of waiting 5-10s for the GPT-4o synthesis. Intent analysis now edits the placeholder. - intent model: gpt-4o → gpt-4o-mini (classification-only, same accuracy for intent extraction at ~3x throughput / lower cost). - progress template: switch from "1-line-per-update" constraint to a fixed section layout (📋 理解 / 🔍 進捗 / 📎 成果物 / ✅ 完了) so users can actually see what the agent is doing on long tasks. - progress model: route small diffs (<2000 chars) to gpt-4o-mini; large diffs stay on gpt-4o. ~70-80% token cost reduction per task. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- PR URL extraction: scan agent output for github.com/.../pull/N URLs and pin them to the 📎 成果物 section so freshly-opened PRs can never fall off the message as the agent generates new findings. - ❌ reaction cancel: subscribe reaction_added events; reacting to the original user message with ❌ aborts the matching task's AbortController. Author-only check prevents bystanders from killing someone else's task. - image vision: download image attachments from the triggering Slack message into ./attachments/ at workspace creation, list them in PROMPT.txt so the Claude agent can open them with the Read tool (vision-capable). Non-image files are skipped. Slack App config: subscribe to reaction_added (reactions:read scope) to enable cancellation. Existing app_mention / message.* events remain. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ress - maxTurns by complexity: intent LLM now also classifies complexity (simple/medium/complex) and the agent gets 40/100/200 turns respectively. Caps runaway cost on simple lookups while leaving room for complex multi-repo work. - content-hash dedup: include sha256(text)[:8] in the dedup key so message edits with new intent re-trigger, but identical retries within 10s still suppress. - message_changed support: flatten Slack's nested edit payload so the rest of the pipeline can treat edits like fresh messages (paired with the content-hash dedup above). - ErrorCollector event-driven: fs.watch (recursive, per-file 500ms debounce) replaces the 10s polling loop. Slow 60s safety-net poll still runs in case the FS layer drops events. - thread context compression: when a thread exceeds 20 messages, keep the latest 15 verbatim and let gpt-4o-mini summarize the older ones (decisions, open questions, named files/PRs, errors). Slashes intent prompt token usage on long threads. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- agent self-updates: instruct the agent in .bot/AGENT.md to call prbot slack update directly on milestones (plan / subtask done / blocker / 30-60s heartbeat). Agent now owns the user-facing message format using the same section structure (📋/🔍/📎/✅). - synthesizer fallback only: bump the GPT-4o synthesizer interval from 10s → 30s. The agent's direct updates are primary; the synthesizer is just a safety net for agents that go quiet on Slack. - follow-up drain test: cover the TransformStream contract slack-bot.ts relies on for piping mid-task Slack messages into a running SDK session — order preservation, single-reader drain, close semantics, release-and-reacquire writers. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Contributor
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Contributor
There was a problem hiding this comment.
Pull request overview
This PR upgrades ComfyPR-Bot’s Slack integration and agent runtime to improve reliability (webhook retries/dedup), responsiveness (fast placeholder + async intent), UX (sectioned progress), and capability (image attachments + ❌ cancel), while also adding a vendored slack-cli subtree and a CLI feedback reporter.
Changes:
- Replace Socket Mode handling with Slack Events API webhook + signature verification and 1h event_id dedup.
- Move agent execution to Claude Agent SDK with per-task user isolation, follow-up message streaming, progress section updates, PR URL surfacing, image downloads, and reaction-based cancellation.
- Add
slack-clisubtree tooling + subtree sync scripts, plus aprbot feedbackcommand posting to Slack.
Reviewed changes
Copilot reviewed 30 out of 35 changed files in this pull request and generated 7 comments.
Show a summary per file
| File | Description |
|---|---|
| scripts/subtree-sync.sh | Adds a helper to pull/push configured git subtrees. |
| package.json | Adds subtree scripts and Claude Agent SDK dependency. |
| bun.lock | Locks Claude Agent SDK and transitive deps. |
| inbox.yaml | Adds inbox file used for “missed” message replay. |
| lib/zChat.ts | Switches z-chat-completion loading to async/dynamic import wrapper. |
| lib/slack/feedback.ts | Adds Slack feedback posting utility + CLI entry. |
| lib/slack-cli/ts/ts-to-iso.ts | Adds TS helper for Slack timestamp → ISO conversion. |
| lib/slack-cli/ts/safe-message.ts | Adds truncation utilities for Slack text/blocks. |
| lib/slack-cli/ts/parse-url.ts | Adds Slack message URL parsing helpers. |
| lib/slack-cli/ts/parse-markdown.ts | Adds Slack mrkdwn → Markdown conversion helpers. |
| lib/slack-cli/ts/index.ts | Exports slack-cli TS utilities. |
| lib/slack-cli/ts/cached-client.ts | Adds a caching proxy wrapper for Slack WebClient calls. |
| lib/slack-cli/src/slack.rs | Introduces direct Slack Web API client in Rust (reqwest). |
| lib/slack-cli/src/main.rs | Adds Rust slack CLI (msgs/news/search/send). |
| lib/slack-cli/src/lib.rs | Adds NAPI bindings for Slack operations. |
| lib/slack-cli/package.json | Adds NAPI build scripts/targets for slack-cli addon. |
| lib/slack-cli/index.js | Adds platform-aware loader for the NAPI binary. |
| lib/slack-cli/index.d.ts | Adds TS declarations for the NAPI exports. |
| lib/slack-cli/bun.lock | Adds lockfile for slack-cli subproject. |
| lib/slack-cli/build.rs | Sets up napi-build in Rust build step. |
| lib/slack-cli/README.md | Documents slack-cli usage/configuration. |
| lib/slack-cli/LICENSE | Adds MIT license for slack-cli subtree. |
| lib/slack-cli/Cargo.toml | Defines Rust crate/bin/deps for slack-cli + NAPI. |
| lib/slack-cli/.gitignore | Ignores build artifacts for slack-cli. |
| lib/slack-cli/.github/workflows/build.yml | Adds CI workflow to build/release multi-target artifacts. |
| bot/taskInputFlow.spec.ts | Adds unit tests for TransformStream “drain” contract used for follow-ups. |
| bot/task-user.ts | Adds per-task Linux user lifecycle management + stale cleanup. |
| bot/spawn-as-user.ts | Adds sudo-based spawner to run agent subprocesses as task users. |
| bot/slack-bot.ts | Major refactor: webhook mode, dedup, placeholder ack, sectioned progress, SDK agent run, cancel, image download, thread summarization, dynamic model selection. |
| bot/index.ts | Loads .env.local with override semantics and logs secret prefixes. |
| bot/error-collector.ts | Changes to fs.watch-driven error detection with slow polling fallback. |
| bot/cli.ts | Adds prbot feedback command to post to Slack feedback channel. |
| CLAUDE.md | Adds PM2 operational docs for running the Slack bot. |
| .bot/AGENT.md | Updates agent guidance for live Slack progress updates and section structure. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Contributor
Author
|
/copilot review |
- image download: prefix saved filename with file.id (or array index) to
prevent collision when two attachments share a name; prefer
url_private_download over url_private; cap downloads at 25MB so an
oversized upload can't blow up the workspace.
- dedup TTL: webhook-event keys already had a 1h TTL, but the per-message
msg-{ts}-{hash} entries were written without one and would accumulate
forever in Nedb+Mongo. Match the 1h TTL.
- task-user cleanup: build activeIds from (thread_ts || ts) so it matches
workspaceId. Threaded long-running tasks no longer get marked stale and
have their isolated Linux user deleted out from under them.
- input drain: release the TransformStream reader lock in a finally block
so cleanup is deterministic and the stream isn't left half-locked.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Contributor
There was a problem hiding this comment.
Pull request overview
This PR modernizes ComfyPR-Bot’s Slack integration and agent execution model, focusing on reliability (webhook retries/dedup), faster user feedback, and operational safety/cost controls while adding supporting tooling (feedback command and Slack CLI subtree).
Changes:
- Switch Slack bot to webhook-based Events API with signature verification and 1h event_id dedup; add faster placeholder ack and sectioned progress updates.
- Migrate agent execution to Anthropic Claude Agent SDK with per-task Linux user isolation, abort-by-❌ reaction, image attachment download for vision, and dynamic maxTurns/model selection.
- Add CLI feedback submission to Slack, plus vendored
lib/slack-clisubtree utilities and build workflows.
Reviewed changes
Copilot reviewed 30 out of 35 changed files in this pull request and generated 11 comments.
Show a summary per file
| File | Description |
|---|---|
| scripts/subtree-sync.sh | Add git subtree pull/push helper for syncing vendored deps. |
| package.json | Add subtree scripts and Claude Agent SDK dependency. |
| bun.lock | Lockfile update for new dependency set. |
| lib/zChat.ts | Change z-chat-completion loading to async dynamic import + init hook. |
| lib/slack/feedback.ts | New Slack feedback poster + channel resolver/auto-create logic. |
| lib/slack-cli/ts/ts-to-iso.ts | Utility for converting Slack ts to ISO. |
| lib/slack-cli/ts/safe-message.ts | Helpers to truncate Slack text/blocks. |
| lib/slack-cli/ts/parse-url.ts | Parse Slack message URLs into channel/ts tuples. |
| lib/slack-cli/ts/parse-markdown.ts | Convert Slack mrkdwn to Markdown (sync/async). |
| lib/slack-cli/ts/index.ts | Barrel exports for TS slack-cli utilities. |
| lib/slack-cli/ts/cached-client.ts | Proxy-based caching wrapper for Slack WebClient calls. |
| lib/slack-cli/src/slack.rs | Rust Slack Web API client via reqwest. |
| lib/slack-cli/src/main.rs | Rust CLI entrypoint implementing msgs/news/search/send. |
| lib/slack-cli/src/lib.rs | NAPI bindings exposing Slack functions to Node. |
| lib/slack-cli/package.json | NAPI package metadata + build scripts/targets. |
| lib/slack-cli/index.js | Runtime loader for platform-specific .node binary. |
| lib/slack-cli/index.d.ts | Generated TS typings for the NAPI module. |
| lib/slack-cli/bun.lock | Nested lockfile for slack-cli subproject. |
| lib/slack-cli/build.rs | NAPI build setup hook. |
| lib/slack-cli/README.md | Documentation for the Slack CLI subproject. |
| lib/slack-cli/LICENSE | MIT license for the Slack CLI subproject. |
| lib/slack-cli/Cargo.toml | Rust crate config (bin + cdylib + deps). |
| lib/slack-cli/.gitignore | Ignore Rust/node build outputs for subproject. |
| lib/slack-cli/.github/workflows/build.yml | CI workflow to build/upload multi-target binaries. |
| inbox.yaml | Rename/initialize inbox tracking file for missed messages. |
| bot/taskInputFlow.spec.ts | Add unit tests for TransformStream drain contract. |
| bot/task-user.ts | Implement per-task Linux user creation/activity tracking/cleanup. |
| bot/spawn-as-user.ts | Spawn Claude Code CLI as per-task user via sudo. |
| bot/slack-bot.ts | Major bot refactor: webhook handler, dedup, quick ack, SDK agent, cancel, vision, summaries, cost controls. |
| bot/index.ts | Load .env.local with override semantics + log secret prefixes. |
| bot/error-collector.ts | Switch error collection to fs.watch + slow polling safety net. |
| bot/cli.ts | Add prbot feedback command posting to Slack. |
| CLAUDE.md | Add PM2 startup/runbook notes for bot operation. |
| .bot/AGENT.md | Instruct agents to drive Slack progress updates via prbot slack update. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Round 2 of review fixes (codex P0 + 2nd Copilot pass on PR #204). Critical - webhook team_id: Slack's Events API places the workspace id on the envelope (payload.team_id), not on the inner event in some payload shapes. zAppMentionEvent and the message-event guard both require team, so app_mentions/DMs delivered via webhook were getting silently dropped — bot 200'd but never processed. Forward team_id onto event. Hardening - env allowlist: don't spread process.env into the Claude agent subprocess. Pass an explicit set (HOME/PATH/locale, GH/OPENAI/NOTION tokens, SLACK_BOT_TOKEN for prbot slack update, MONGODB_URI, log flags). Slack signing/socket secrets and other unrelated creds no longer leak into bash-tool reach. - follow-up drain buffer: hold values that arrive before agentQuery is ready in earlyBuffer, then flush in order once the SDK is up. Sets agentReady right after query() so we don't drop early follow-ups. - env log prefix: 12 → 6 chars (still enough for an operator to verify the right secret loaded, fewer characters available if the log leaks). Feedback (lib/slack/feedback.ts) — codex P2 + Copilot - Lazy env read: feedbackChannel() resolves PRBOT_FEEDBACK_CHANNEL on call so .env.local override applied inside the CLI handler is honored. - Channel ID regex: accept C/G/D prefixes (was C-only), so private channel IDs and DM IDs pass through without an unnecessary list/create. - Pagination: walk response_metadata.next_cursor to find channels past the first 1000. - Fail closed: never auto-create the feedback channel. Earlier code created it as is_private:false, which would leak the very first submission (stack traces, command output) into a brand new public channel if the env var was misconfigured. Now we throw a clear setup error instead. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Contributor
Author
|
Addressed in commits 1e89d22 and 3d0cbdf: Critical (codex P0)
Copilot 1st-pass fixes
Copilot 2nd-pass fixes
Feedback safety (codex P2 + Copilot)
Deferred (out of scope for this PR / pre-existing)
|
- prbot slack read @snomiao → resolve user via users.list pagination, open DM via conversations.open, then read recent messages. - gitignore: add local tool artifacts (.codex, .playwright-mcp/, .playwright-cli-multi-tab/) so rech/codex output doesn't show up as untracked noise on every git status. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Rebase was tried first but failed because the slack-cli subtree commits on this branch are duplicated by PR #187 (already on main); a merge is much safer and produces the same end state for PR #204. Conflict resolutions - CLAUDE.md: keep main's createIndex security note alongside our PM2 bot startup section. - bot/slack-bot.ts: keep ours. Our version is a superset — webhook mode, sectioned progress, dedup, image vision, ❌-cancel, follow-up drain, dynamic maxTurns — already includes the throttle/extract cleanup that main's e0abd3e introduced. - lib/slack/feedback.ts: keep ours. Our version has the privacy fix (fail-closed instead of auto-creating a public channel) and lazy env read for PRBOT_FEEDBACK_CHANNEL. - bun.lock: take main's; lockfile is downstream of package.json which has only additive merges. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
🎉 This PR is included in version 1.5.0 🎉 The release is available on: Your semantic-release bot 📦🚀 |
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
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
ComfyPR-Bot behavior improvements (16 items total), shipped in phases across Slack webhook, UX, cost, features, and reliability.
Reliability (Phase 1)
event_iddedup)..env.localoverride + secret-prefix log on startup — prevent the recurring incident (2026-04-24) where pm2 carried stale shell env that shadowed.env.local.UX / response quality (Phase 2)
Feature additions (Phase 3)
github.com/.../pull/Nfrom agent output and pin them under 📎 Deliverables../attachments/so the agent can read them via the Read tool.Cost / reliability finalization (Phase 4)
event.ts + sha256(text)[:8]so an edit with new intent retriggers, but identical retries within 10s do not.bun run checkintest.yaml)..bot/AGENT.mddelegatesprbot slack updateto the agent itself. The GPT-4o synthesizer is demoted to a 30s safety net.Test plan
bunx tsc --noEmit)bun test bot/taskInputFlow.spec.ts(4/4 pass)Ignoring duplicate webhook eventlog[env] SLACK_SIGNING_SECRET=f944ab... | ...Design notes in
tmp/improvement-plan.md(gitignored).🤖 Generated with Claude Code