[upstream-take #7219] fix(shutdown): kill detached agent processes on server shutdown (#7218)#176
Merged
AnilChinchawaleXDC merged 27 commits intoJun 1, 2026
Conversation
Co-Authored-By: Paperclip <noreply@paperclip.ing>
When running in Paperclip, interactive permission prompts cannot be answered, making it critical for all executions (including subagents) to have permissions pre-approved via --dangerously-skip-permissions. This fix forces dangerouslySkipPermissions=true when executing within a Paperclip context (detected via context.paperclipWorkspace), ensuring that: 1. Subagents spawned by the Agent tool inherit Bash and other permissions 2. Main session permission settings propagate to all child executions 3. Non-interactive runs never get blocked on permission approval prompts Fixes: MVA-392 (subagent Bash permission denied in /team-implement) Co-Authored-By: Paperclip <noreply@paperclip.ing>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…exhausted Add a pre-flight quota check to `tickTimers` that calls `fetchAllQuotaWindows()` before dispatching any timer-based agent wakeups. If any window reports ≥95% usage, all timer wakes in that tick are skipped and a structured warn is logged with the reset timestamp. Fails open: a quota-check error is logged but does not block wakeups. The `tickTimers` return value now includes `quotaBlocked` and `quotaResetAt` fields, and `index.ts` logs a distinct `quota_blocked` warn when they are set. Co-Authored-By: Paperclip <noreply@paperclip.ing>
Extend the final-disposition checklist in the Paperclip skill to explicitly require removal of any git worktrees and temp clones created during the work before the issue can be set to `done`. Adds concrete commands for both the AutoBot worktree path and the Paperclip dev-clone path. Co-Authored-By: Paperclip <noreply@paperclip.ing>
…pt logging Agents must now: - Post a structured "Work artifact" comment whenever they open a PR, push commits, or create a branch - Log every fix attempt outcome (success or failure) before exiting the heartbeat, including approach taken, error output, and next step Also documents the periodic orphaned-work sweep the CEO should run: - Unlinked open PRs (branch name matches issue-NNNN but no active Paperclip issue) - Stale worktrees whose issue is done/cancelled - Dead in_review issues whose linked PR was already merged or closed Co-Authored-By: Paperclip <noreply@paperclip.ing>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…tine" This reverts commit 5427933.
…ager routine"" This reverts commit c896302.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
… meta-project Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Onboarding now reads README, docs, dependency manifests, env templates, and CI config to build a structured inventory of what the project needs to run, then creates setup tasks from any gaps found. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Onboarding now sweeps open issues (classifying critical vs backlog) and open PRs (failing CI, stale approved, needs review) and creates Paperclip tasks for each actionable item found. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…kickoff Every discovery item (setup gaps, docs gaps, critical GH issues, PR actions) becomes a child issue of the kickoff issue, grouped under category parent issues. Kickoff auto-closes when all children are done. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Every created issue must have a label and correct projectId. A sweep check runs before moving past Step 3. Labels also applied back to GH where missing (e.g. unlabelled bugs get the bug label). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Every project gets a permanent [Docs] hub issue with 5 required documents: prd, tech-stack, access-guide, architecture, runbooks. Populated from repo discovery; unknown sections stubbed as _Not yet documented_. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
User creates empty project with repo URL → PM detects no docs hub → runs full onboarding: reads repo, extracts content into 5 hub documents, structures all findings as sub-issue tree, labels everything. Content is extracted from existing repo docs, not just stubbed. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
6-task plan: GH labels, dispatch AGENTS.md section, 30-min routine, onboarding AGENTS.md section, dispatch smoke test, onboarding smoke test. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Co-Authored-By: Paperclip <noreply@paperclip.ing>
…rclipai#7218) Export signalRunningProcess from adapter-utils and call it for all runningProcesses entries in shutdownAppServices(). Previously, agent subprocesses spawned with detached:true escaped the systemd cgroup and survived server restarts, accumulating as orphans that consumed memory and left routine execution issues stuck in_progress.
Write oom_score_adj=500 to /proc/<pid>/oom_score_adj immediately after spawning each detached agent subprocess. This makes the kernel prefer killing agent processes over the Paperclip server when memory is tight, preventing OOM-kill of the server and embedded postgres mid-query. Companion to systemd OOMScoreAdjust=-500 on the service side.
Adds the same Claude Code discipline used in AutoBot: - .claude/hooks/: block-dangerous-commands.sh (28 tests pass), protect-files.sh, scan-secrets.sh — all adapted for paperclip's master branch + pnpm stack - .claude/settings.json: wires hooks to PreToolUse/PostToolUse/Notification events; PostToolUse runs pnpm -r typecheck on .ts/.tsx edits - .claude/skills/: commit, bugfix, implement, research, parallel — all rewritten for paperclip's branch strategy, AGENTS.md contracts, and pnpm commands Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
… PID When the server crashes/restarts before persistRunProcessMetadata writes the child PID to the database, reapOrphanedRuns() cannot check whether the process is still alive. Previously it immediately treated the run as dead, called releaseIssueExecutionAndPromote + startNextQueuedRunForAgent, and dispatched a replacement — while the original process kept running as an untracked orphan. On each subsequent restart this repeated, leading to exponential orphan accumulation that exhausted the WSL memory cap and froze the host. Fix: when tracksLocalChild=true but both processPid and processGroupId are null, skip the run if it was updated within the last 10 minutes. The slot stays marked "running" and countRunningRunsForAgent() counts it, so the concurrency cap remains enforced. After the grace window the slot is reclaimed as before. The startup reaper (staleThresholdMs=0) and the periodic reaper (5-min threshold) both benefit. Two new tests cover the grace-period boundary: - recent null-PID run: reaped=0, status=running (grace period active) - old null-PID run: reaped=1, status=failed (grace period expired) All 46 process-recovery tests pass. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…m-pr-7219-kill-detached-agents-on-shutdown
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
3 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.
No description provided.