feat: claim_work MCP tool for session dedup#14
Merged
Conversation
channelInfoCache in http-routes.ts was never populated for non-DM channels, so most sessions showed no channel name. Now resolves channel info via Slack API (conversations.info) with caching, used by both main dashboard and personal dashboard. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Extract artifacts (PRs, issues, gists) and latest reply from activity.jsonl in buildDashboardData. Render as compact clickable chips and truncated reply preview on each workspace card. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
All dashboard entries now use ThreadCard (collapsible). Collapsed view shows status dot, artifacts inline in header, and reply preview below. Running sessions auto-expand. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ck msg - Default profile: fallbackRef=origin/next, description tells Claude to use next - Claudebox-dev: fallbackRef=origin/main - Soften fallback message from WARNING to informational note Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Make it clear that session_status should be called after every major step, with examples. Users watch progress live and see nothing without these calls. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Double-RAF + 100ms mount scroll ensures content is painted before scrolling. Fixes page not starting at bottom for long activity feeds. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Bot must call session_status before anything else so the user sees immediate feedback. Even for trivial prompts like "wake up", post status so the user knows the bot is alive. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…vements - Add container recovery on server restart — re-attach streamers and exit monitors to Docker containers that survived the restart - Fix massive activity.jsonl bloat from cumulative progress events — deduplicate tool_use entries by ID in session streamer, scope subagent discovery to current session only, add server-side dedup on SSE init - Add `claudebox creds` CLI for managing multiple OAuth credential slots (list/use/save/rm) stored in ~/.claudebox/credentials/ - Add Agent Team UI component for rendering parallel agents on status page - Fold tool results by default on status page (click to expand) - Fix literal \n in GitHub comments — unescape in all body/content parameters - Fix parallel tool call visibility — increase trunc limit, add timestamp to msgId - Fix activity bleeding across runs — respect entry's own log_id in SSE handler - Optimize SSE polling with byte-offset incremental reads - Add set_workspace_name to all profile workflows (prevents ugly branch names) - Remove update_meta_issue tool and all meta-issue references (use gists instead) - Delete minimal profile - Strengthen set_workspace_name tool description to MANDATORY Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
… timer - Recognize Slack URLs in link field and show as Slack origin with ↗ icon - Promote link to slackLink when session has Slack URL but no slack_channel - Add hourly OAuth token refresh via systemd timer Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Show elapsed time since session start in every session_status update - Include elapsed in GitHub completion comment (e.g. "✅ Run #1 — completed (12m)") - Include elapsed in Slack status footer - Compute from session.started or first activity timestamp Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- app_mention handler now skips bot_id and subtype messages - Replace regex with indexOf for Slack URL check in dashboard template literal Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…at hooks - Replace ci.sh shell-out in read-log with direct redis-cli + gunzip (fixes .fzf.bash pollution) - Fix redis empty-key detection (strip trailing newline before length check) - Fix HTTP fallback to use Basic auth header instead of URL credentials - Add comprehensive make target listing to build tool description - Add yarn_project_format and barretenberg_format MCP tools - Add formatBeforePush hook to PR creation/update flow - Mount reference repo node_modules for prettier in containers - Add build-tools.ts, bootstrap.sh, install.sh, claudebox CLI - Update barretenberg-audit profile and Dockerfile Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add review profile with comprehensive aztec-packages review knowledge base - Extract common aztec-packages sidecar setup into profiles/shared/ - Add GitHub webhook endpoint for claude-review label trigger - Add manage_review_labels MCP tool and removeLabel to libcreds - Move SESSIONS_DIR from .claude/ (credential-scoped) to ~/.claudebox/ to survive credential changes — fixes dashboard showing no sessions - Add elapsed time to status comments - Direct redis read_log (no ci.sh dependency) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Fixes CodeQL actions/missing-workflow-permissions alerts. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ensitive information Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Build tools (build, build_cpp, run_test, yarn_project_format, barretenberg_format) were causing sessions to get stuck. Replaced with documentation in CLAUDE.md pointing to make targets, bootstrap.sh fallback, and cmake commands. Formatting still runs automatically before create_pr/update_pr via the formatChangedFiles hook. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…session When a new session starts on an existing worktree, any containers from previous runs (claude + sidecar + network) are now stopped and removed first. This prevents stale sidecars with old MCP tools from lingering. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- POST /run now accepts slack_channel and slack_thread_ts; when provided, posts an initial Slack message and updates it on completion, making HTTP-initiated sessions behave like Slack-originated ones - Remove auto-resume on queue drain — users must explicitly reply - Remove auto-resume in reconcile — always mark orphaned sessions cancelled - Auto-merge failure detection triggers sessions from bot messages - Relax merge-base check to warning instead of blocking PR creation Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…to-resume - POST /run now resolves slack_channel/slack_thread_ts from the resumed session (via PR binding or worktree_id). When found, posts a reply in the original Slack thread and updates it on completion — no workflow changes needed, works out of the box with existing claudebox.yml - Also accepts explicit slack_channel/slack_thread_ts in body as fallback - Remove auto-resume on queue drain — users must explicitly reply - Remove auto-resume in reconcile — always mark orphaned sessions cancelled - Auto-merge failure detection triggers sessions from bot messages - Relax merge-base check to warning instead of blocking PR creation Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When /run has no Slack context from a prior session, reverse-lookup the channel from the channelProfiles map using the session's profile. This means GitHub-triggered sessions (issue_comment, workflow_dispatch) post a status message in the profile's Slack channel automatically — no workflow changes needed. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Instead of reverse-looking up profile→channel, parse the Slack channel and thread_ts directly from the link field when it's a Slack URL (e.g. https://workspace.slack.com/archives/CHANNEL/pTS?thread_ts=...). This works out of the box — callers just pass the Slack URL as the link. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
… skill - Force all gists to be private (remove public_gist param from create_gist tool, hardcode public:false in createGist API) - Add GCP service account support: bind-mount SA key into containers, activate in container-entrypoint.sh before Claude starts - Migrate container HOME from /home/claude to /home/aztec-dev to match the devbox image user (uid 30079). Updated all fallback paths in mcp/server.ts, mcp/env.ts, cli.ts, tests, and settings. - Rebuild audit-devbox from local devbox:latest to align uid - Add network-logs skill and agent to default profile for GCP log queries - Update CLAUDE.md with GCP instructions and build guidance (no Docker) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Agents must call claim_work() as their first action. It atomically records what they're about to do and returns all sessions from the last 24h. The agent semantically analyzes for overlap — if another session is already handling the same task, it responds with the existing session's log URL and exits. - POST /api/internal/claim-work endpoint on host server - claim_work MCP tool in sidecar (calls host, returns recent sessions) - HostClient.claimWork() method - Updated CLAUDE.md workflow to mandate claim_work before any work Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Bind-mount ~/.bb-crs into both sidecar and Claude containers at $HOME/.bb-crs:ro, matching the CI bootstrap_ec2 pattern. Avoids 2GB re-download per session. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
readBody returns a string, not parsed JSON. Destructuring without JSON.parse gave undefined for all fields. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Only compare against sessions with the same profile — audit sessions shouldn't dedup against default sessions and vice versa. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- "stop", "cancel", "kill", "abort" in a running session thread now kills the container immediately instead of queuing the message - claim_work prompt is much stricter: same PR = same work even across branches (v4-next vs backport-to-v4-next-staging), automation sessions are flagged as duplicate-suspect, no "proceed anyway" escape hatch unless prompt explicitly says "ignore duplicates" or "force" Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- All draft PRs created by claudebox now get both 'claudebox' and 'ci-draft' labels automatically - Removed "ignore duplicates" / "force" escape hatch from claim_work — if there's overlap with a running session, agent MUST exit, no exceptions Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When overlap is detected, agent now asks the user in Slack whether to proceed or let the existing session finish. User can reply to restart. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Only create issues for genuinely new findings. If a finding already has a linked GitHub issue, it is not new — skip it. Co-Authored-By: Claude Opus 4.6 (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
2 participants
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
claim_work()as their first action, describing what they're about to doChanges
POST /api/internal/claim-workendpoint on host serverclaim_workMCP tool in sidecarHostClient.claimWork()methodclaim_workbefore any workTest plan
🤖 Generated with Claude Code