Paperclip
Paperclip signs workspace handoffs before declaring readiness
Paperclip proposes signing workspace login handoffs and gating readiness, alongside database-seed verification and an atomic drain fence for agent execution. The three changes move workspace startup toward a verifiable state transition.
paperclipai/paperclip PR #11671 is the inspected primary source: “feat(workspaces): sign the workspace login handoff and gate readiness.” The related records below were inspected as supporting context rather than independent confirmation.
The facts
- paperclipai/paperclip PR #11671 was created 2026-08-19T00:51:55Z and was open at inspection; its title is “feat(workspaces): sign the workspace login handoff and gate readiness.” - paperclipai/paperclip PR #11665 was created 2026-08-18T23:20:33Z and was open at inspection; its title is “feat(workspaces): verify database seeds and add managed repair.” - paperclipai/paperclip PR #11664 was created 2026-08-18T23:13:12Z and was open at inspection; its title is “feat: promote atomic agent execution drain fence to upstream.” - The repository reported 78813 stars, 5159 open issues, default branch master, and last push 2026-08-19T01:04:10Z. - The latest tagged-release baseline checked was v2026.817.0 published 2026-08-18T03:17:29Z.
What changed
paperclipai/paperclip PR #11671 reports: Thinking Path - Paperclip is the open source app people use to manage AI agents for work - Managed git-worktree workspaces each run an isolated Paperclip instance with its own cloned database, published over HTTPS - Those workspace URLs reported status: ok and bootstrapStatus: ready, then answered the operator's sign-in with "invalid email or password" — the workspace was live, but the only way in was a password captured when the clone was made - Transport health cannot detect this. A workspace can answer HTTP 200 while its clone has no usable rows, its own instance no longer owns the port, or its login path depends on a credential nobody remembers - This pull request adds a signed single-use login handoff so entry is password-independent, and a protected readiness contract so a workspace cannot publish itself as healthy unless its database, cloned data, login handoff and instance identity all agree - The benefit is that opening a workspace works without remembering a snapshot password, and a broken workspace says which phase failed and what one action fixes it instead of showing a green badge or a generic load failure Linked Issues or Issue Description No public GitHub issue exists for this work, so the problem is described here. What happened** Two managed workspace URLs on separate worktrees both returned HTTP
paperclipai/paperclip PR #11665 reports: Thinking Path - Paperclip is the open source app people use to manage AI agents for work - Isolated workspaces clone control-plane data into a separate database - A binary seed marker did not prove that restore, migration, quarantine, and validation completed - A damaged workspace also had no bounded database-only recovery operation - This pull request adds verified seed evidence and one managed repair lane - The benefit is that a partial clone cannot start and an operator can recover a workspace without changing its files Linked Issues or Issue Description Refs #11651 What happened?** Worktree database startup trusted binary marker files. A partial restore could leave weak readiness evidence. Operators had no audited operation that stopped services, backed up the isolated database, reseeded it, and proved readiness before restart. Expected behavior** Paperclip must start an isolated workspace only after one explicit validation contract passes. A repair must preserve worktree files, retain recovery evidence, and report the exact failed phase. Steps to reproduce** 1. Interrupt a worktree database seed during restore or migration. 2. Start the worktree runtime. 3. Observe that the old binary marker model cannot describe the failed phase or validate the cloned auth and company data. Paperclip version or commit** Current master plus the workspace lifecycle work in #11651. Deployment mode** Local development with managed git
paperclipai/paperclip PR #11664 reports: Thinking Path - Paperclip is the open source control plane that companies use to manage AI agents at work. - The heartbeat service admits, executes, and finalizes agent runs for each company. - A run can overlap a pause or a drain decision, which lets a paused agent keep executing work. - The drain fence makes pause and drain atomic: it blocks new execution work while a fence is held and certifies finalization only after the execution process is gone. - Portable recovery and terminal acknowledgment handling keep the fence correct across platform and crash cases. - On 64-bit Windows, a 32-bit Node process could not inspect a native 64-bit target; the Sysnative redirect fixes that. - This pull request promotes the reviewed drain-fence work onto current upstream master and opens it as a fork-to-upstream change. - The benefit is a verifiable, atomic execution drain that pause and drain operations can rely on in production. Linked Issues or Issue Description Bug:** The drain fence is missing from upstream master. Expected behavior:** Pause and drain must atomically stop agent execution. Actual behavior:** Without the fence, a paused agent can continue executing work. Impact:** Pause and drain are not atomic in the current release. Steps to reproduce:** Acquire a drain while a run is admitted. What Changed Added the atomic agent-execution drain
Why it matters
A workspace is not ready because a process answers. Identity handoff, seed integrity and in-flight execution all need to settle before work is admitted. The operator test is whether the system remains bounded and its receipts still describe the action after failure, retry or restart.
Current
The primary record was open when captured on 2026-08-19. The tagged-release baseline was v2026.817.0 published 2026-08-18T03:17:29Z. Repository metadata, full source bodies, current pull requests and issues, releases, Google News, Hacker News, Lobsters, Metamesh and the rendered ClawCharts row were inspected. ClawCharts selected the subject; it did not prove the claim.
Evidence
The primary URL is https://github.com/paperclipai/paperclip/pull/11671. Supporting records are https://github.com/paperclipai/paperclip/pull/11665 and https://github.com/paperclipai/paperclip/pull/11664. Source bodies, timestamps and states are preserved in the daily evidence bundle. Test counts and reproductions remain attributed to their authors unless identified as independently rerun.
Source boundary
Open work is described as open, closed work as closed, and operator reports as reports. A pull request is evidence of proposed or reviewed direction, not proof of a shipped release. Search residue, package mirrors and historically published source spines were excluded.
Operator take
Bind login handoffs to signed state, verify seeded data before readiness, and drain agent execution behind one atomic admission fence. Preserve a before-state receipt, make the smallest reversible change, and verify the original failure independently.
Caveat
Public project records are mutable. Status, scope and evidence can change after publication. This brief records the inspected state and does not authorize changes to a reader’s deployment.
Source inspected; source state, environment and release boundary remain explicit.