## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Native runs must preserve files when Stop ends a turn before its result is saved. > - The cancellation path could skip file retention and leave a prepared remote workspace active. > - An older controller could also admit an extra run while an earlier lost-source hold was unresolved. > - Rejecting every later run then prevented recovery even after that extra run was stopped and its files were reconciled. > - This change retains cancelled workspaces and records the exact proof for a safely reconciled intervening cancellation. > - Recovery can proceed without changing cancelled history, replaying actions, or weakening execution gates. ## Linked Issues or Issue Description Refs: #16583, #16690, #16696. **What happened?** Stop could acknowledge a native turn without recording its unsaved file obligation. Reconstruction also rejected a later cancelled run after its exact environment and files had been reconciled. The task could remain inactive with an `in_progress` status, which the reconstruction endpoint rejected. **Expected behavior** Retain unsaved files after Stop. Permit a board operator to reconstruct from saved sources only after each intervening cancellation has a confirmed stop and a file-only repair record. Preserve the original failed export and the cancelled run. **Steps to reproduce** 1. Prepare a native workspace and cancel the turn before a structured result exists. 2. Inspect its file retention and remote lease state. 3. For a task with an older destroyed export, reconcile the later cancelled run's exact files and unperformed actions. 4. Request reconstruction from verified same-task attachments. ## What Changed - Pin prepared native files before cancellation settlement. Defer teardown if pinning fails. Use the existing stop-only cleanup worker. - Repair active warm and ephemeral sources, plus stopping and confirmed-stopped historical cancellations in bounded startup and periodic batches. Preserve the existing exact-source checks. Bound row-lock waits and advance past a busy source. - Accept only same-owner native run-only cancellations with no result or current coordinator, confirmed stopped leases, and board file-only repair with `not_performed` action outcomes. - Save hashes of each intervening repair and recheck them at queue claim and final dispatch. Reject changed evidence or another later run. - Expose saved-source reconstruction in the issue API for inactive `blocked` and `in_progress` tasks, and hide it when an execution or checkout owner exists. Permit the same inactive source at submission. Keep the source, owner, version, attachment, budget, pause and retry checks. - Update the shared evidence type, recovery runbook and capability maps. Add an optional cancelled-native no-action confirmation to file recovery, resetting evidence when its source changes. No migration, new endpoint, permission or dependency. ## Verification - The new fixture fails on the original validator with `workspace_reconstruction_stale` after valid cancelled-run file repair. - Focused source and Stop retention tests: 94 passed before the final contention case. The subsequent three-suite run passed 123 tests, including real PostgreSQL lock contention and public startup/API/final-dispatch paths. The final narrow-query rerun passed all 123 tests. Review correction: all 143 tests passed across four suites, including real lock contention through the teardown boundary and stopping/stopped allocations. - Six broader cancellation, executor, heartbeat and restart suites: 831 passed, 9 existing skips. - Final correction: all 145 backend tests and all 36 recovery UI tests passed. Real contention covers both warm reusable and ephemeral sources. Optional no-action confirmation defaults to uncertified, and changing source clears evidence. Workspace `pnpm -r typecheck` and token gates passed on these sources. - Maps, module boundaries and Git whitespace checks: passed. Full `pnpm build` passed on the final `4e8c0e73ac` revision as well as the prior reviewed heads. A full local run of the preceding fix completed with 21,724 passes, 109 skips and three failures in webhook body limits, a status-card timeout, and a port-allocation fixture. All three suites then passed unchanged in an isolated rerun (209 tests). The full local run remains recorded as failed; the final head's full CI passed (55 completed checks, no failures). A fresh full local run is also in progress on the frozen reviewed head. - Preserved failures: the first attempt lacked a built plugin SDK; after building prerequisites, the regression reached the expected validator rejection. Treatment tests caught the file-only action's deliberately null continuation owner. A real lock test then caught the database driver's wrapped timeout error. Both defects were corrected and retested. Independent review then found a teardown race after a failed retention write. A second review found active warm sources excluded by the historical policy filter; the regression reproduced both warm failures before correcting the query. Its real-row-lock fixture reproduced the failure; the fixed release guard and historical stopping/stopped cases pass. No checks were weakened. - A pre-deployment walkthrough caught the issue API still hiding reconstruction for an inactive `in_progress` task. The status/capability baseline reproduced three failures, including inappropriate capability flags for execution/checkout owners. The fix uses the same inactive-state requirements as submission; its positive fixture reads the assembled issue API before reconstructing. All 117 tests in those three recovery suites pass, and all 28 heartbeat-recovery tests pass separately. Full workspace typecheck passed again. Full build passed with the exact `4e8c0e73ac` stamp; latest-head independent review is 5/5 with no open findings. All 55 latest-head checks completed with no failures. - Live tenant acceptance remains pending. No successful reconstruction, export or warm-process proof is claimed by these local tests. ## Risks - Stop retains unfinished work, so storage remains until file recovery and explicit cleanup complete. - Reconstruction remains held for unknown or performed external actions, foreign owners, live or rebound environments, accepted results, changed evidence, and unrelated later runs. - The sweep examines at most five candidates per pass. It uses a fair cursor and bounded transaction waits. It performs no provider call and starts no new turn. - Intervening verification reads only the needed JSON fields. The source transaction still serializes recovery for the same task. ## Model Used OpenAI Codex, GPT-6 family, with reasoning, tool use and code execution. The session does not expose an exact backend model ID or context-window size. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>
14 KiB
Execution workspaces, services, and files — agent access
Product map: Execution workspaces, services, and files.
Lost-source reconstruction adds the board-only
POST /api/issues/{id}/recovery-actions/reconstruct-workspace action. It requires
runtime:manage and ordinary task visibility. Agents cannot authorize it. The
result is a durable fresh-session admission request, subject to current execution
gates. It grants no additional tool access, budget, work window or permission to
repeat completed actions. Attachments must belong to the same task and company.
The original failed run and lost-export evidence remain unchanged.
Later native run-only cancellations can coexist only with confirmed lease stops,
board file-only repair and an external-action outcome of not_performed.
Their immutable evidence is checked again at dispatch. Historical cancellation
retention runs internally and grants no new agent operation or authority.
Review date: 2026-10-12. Source commit: 58e16dbddfc9c8d0cf60867f9d9847f0feda055b.
Native REST transport reconciled in this change; see the
shared transport contract.
This is a source-reviewed baseline of representative operations and known gaps. Each capability needs operation-level review before its entire scope can be qualified. No live agent execution is claimed. Follow the map contract for format, permission checks, evidence and maintenance.
binding
Review: baseline.
API: present — representative operation IDs:
GET /api/issues/{id}/heartbeat-context.GET /api/execution-workspaces/{id}.GET /api/issues/{id}/workspaceinspects the durable task binding.PUT /api/issues/{id}/workspacerequests the next binding.GET /api/companies/{companyId}/execution-workspaces?summary=true&selectableForTask=truelists authorized task-selectable workspaces.
Agent access: available — Native get_task_context / REST / CLI / MCP workspace read.
The task-bound native tools get_workspace, list_workspaces, and
select_workspace use the ordinary authenticated APIs. Legacy agents can call
those APIs with their admitted run token. These paths do not require a project.
Policy: Company/project/workspace visibility and runtime authority; active run, trust preset and task ownership where required.
Task workspace writes require issue:mutate, an active owned task/run for agents,
and current source visibility. Shared physical roots also require source-project
workspace write authority. Accepting a new pending choice advances its binding
revision; superseded retries fail while current-choice retries remain idempotent.
Binding revisions and request keys fence retries;
selection applies at the next normal admission without interrupting a live run.
Fresh standard-trust tasks with an implicit repository default can use an owned
remote clone when the shared default source fails the restricted-task workspace
guard. This does not grant access to the private local checkout or its Git metadata.
The normal admitted Git identity must already authorize the configured remote.
Creation of a fallback source is unavailable for explicit, parent/channel, reused
or low-trust selections, or branch templates without a task-derived placeholder;
an existing verified owned source remains reusable. Literal branch selections
retain the original denial without cloning or replacing their unpublished commits.
New implicit Git defaults record project-source provenance. Untyped explicit,
inherited and older ambiguous checkout bindings retain the denial; admission reads
the current company-scoped binding before considering the fallback.
A retained denial blocks the task as configuration_incomplete
(workspace_source_restricted) before dispatch; it grants no access and names the
owner action on the task.
Local paths, unsupported transports and credential-bearing URLs are not fallback
sources. Company/task/remote bindings, real paths, Git storage and the exact origin
are revalidated for reuse, cleanup and reopening. A failed save removes only the
owned task worktree. The isolated-source service
and its local regression tests establish source and lifecycle boundaries; live
agent/provider qualification remains separate.
Reconstructable Git expiration adds no agent endpoint or permission. It records
company/workspace/repository identity and an exact commit before removing owned
cache files. Existing task/workspace reopening retains current company, source,
and trust checks. Missing upstream access returns a retryable error; it does not
choose a newer commit or grant access to another company's source. Shared project
bases and unresolved recovery remain protected. Local lifecycle evidence is in
expiration tests.
Writable aliases must also be compatible with current and retained physical source projects' trust boundaries. Incompatible choices are rejected before dispatch; a projectless alias cannot bypass required sandbox containment. Source-copy checks apply the same trust boundary before provisioning. The authoritative sandbox or SSH selection suppresses configured and built-in controller worktree and runtime provisioning regardless of task trust, source write grants or board sponsorship. Genuine isolated worktrees still permit authorized read-only source copying. Source write permission does not authorize controller execution. Explicit, channel, inherited, foreign or nondefault isolated source choices remain copy-only for controller provisioning even on local runs. Their deny-only workspace marker survives metadata and binding changes; a board reopen does not turn it into approval. Implicit trusted local defaults and explicit SSH remote provisioning retain their existing placement; no new sandbox provision transport is claimed. The implicit default follows task creation: configured policy pointer, then primary source, creation time and ID. A missing or foreign configured source does not authorize controller hooks. This precedence neither changes existing creation validation nor rotates bindings or clears retained copy-only markers. Closed-worktree reopening checks current task, agent, organizational and source policies without assuming a future driver. Effective low-trust reopens suppress configured and built-in host scripts. Returned checkpoints must have directory parents for every nested entry before merging, including unchanged entries; invalid output leaves retained host files intact and keeps recovery evidence.
Skipping the host scan for a fully unchanged returned snapshot grants no new access. The merge still holds its canonical writer lock and runs both authority and publication hooks. Denials still propagate, and checkpoint/archive validation is unchanged. Regression coverage is source-level evidence only. Child creation by an admitted agent checks an explicit prospective-task context before the child is saved. The creator run must have an authoritative saved policy and a verified company/agent/parent lineage. Current parent, prospective child and physical source policies must fit that boundary. This creation path does not relax exact task/run checks for existing-task writes. The child trust regression is automated source-level evidence, not live agent qualification. Source review for these additions: PR #15749 and task workspace routes, native authority, and semantic transport.
Evidence: source-only —
implementation at the
reviewed commit above.
Verification: not-run — exercise a listed permitted operation with actual agent
credentials/run context; inspect its result and applicable persistence/attribution.
Assert the relevant denied actor, owner, company or mode from Policy. Review the
remaining operations before qualifying the whole capability.
Gaps: Run/task context identifies its actual realized workspace; record existence is not a running service.
provisioning
Review: baseline.
API: present — representative operation IDs:
GET /api/execution-workspaces/{id}.GET /api/execution-workspaces/{id}/workspace-operations.POST /api/issues/{id}/workspace/repositoriesrequests repository preparation.
Agent access: partial — REST / native call_api reads; task/project assignment triggers
server provisioning.
Native prepare_repository and admitted legacy REST callers can request a
catalog repository or validated repository URL for their current task.
Preparation occurs at normal admission; the agent cannot supply a host path or
controller identity. Current source permissions, responsible-user intersection,
trust restrictions, immutable receipts, and retained physical-root ownership are
checked before acquisition and publication. Low-trust callers cannot trigger a
new controller-side Git acquisition. Evidence remains source-only here; focused
repository tests
are separate from live agent qualification.
Policy: Company/project/workspace visibility and runtime authority; active run, trust preset and task ownership where required.
Evidence: source-only —
implementation at the
reviewed commit above.
Verification: not-run — exercise a listed permitted operation with actual agent
credentials/run context; inspect its result and applicable persistence/attribution.
Assert the relevant denied actor, owner, company or mode from Policy. Review the
remaining operations before qualifying the whole capability.
Gaps: Observing readiness exists. Creating host commands or arbitrary provision targets is a separate protected operation.
services
Review: baseline.
API: present — representative operation IDs:
POST /api/execution-workspaces/{id}/runtime-services/{action}.
Agent access: available — Authorized agent REST / native call_api / CLI runtime
service controls and existing stdio MCP controls.
Policy: Company/project/workspace visibility and runtime authority; active run, trust preset and task ownership where required.
Evidence: source-only —
implementation at the
reviewed commit above.
Verification: not-run — exercise a listed permitted operation with actual agent
credentials/run context; inspect its result and applicable persistence/attribution.
Assert the relevant denied actor, owner, company or mode from Policy. Review the
remaining operations before qualifying the whole capability.
Gaps: No dedicated production control_workspace_service binding. Generic native calls retain runtime trust and workspace/task scope checks. Optional MCP availability must be recorded.
files-git
Review: baseline.
API: present — representative operation IDs:
GET /api/issues/{issueId}/file-resources/content.GET /api/issues/{issueId}/file-resources/list.
Agent access: partial — Authorized workspace filesystem tools; Dot workspace bridge
when enabled; board file-resource HTTP UI.
Policy: Company/project/workspace visibility and runtime authority; active run, trust preset and task ownership where required.
Evidence: source-only —
implementation at the
reviewed commit above.
Verification: not-run — exercise a listed permitted operation with actual agent
credentials/run context; inspect its result and applicable persistence/attribution.
Assert the relevant denied actor, owner, company or mode from Policy. Review the
remaining operations before qualifying the whole capability.
Gaps: Files panel HTTP requires board. Agents can use their admitted workspace file tools; remote/Dot scope and Git command availability differ.
File reads also enforce retained source privacy after local copy, configuration move/removal, reopen, and project deletion. A task assignment or repository receipt does not grant access to private files from another physical owner.
Authorized Git-source capture also retains ownership on the actual shared Git storage. Source checks use verified checkout identities, including company-known owners when separate metadata has no primary-checkout registration. Removing that checkout does not release its private committed files in surviving worktrees.
closure
Review: baseline.
API: present — representative operation IDs:
PATCH /api/execution-workspaces/{id}.GET /api/execution-workspaces/{id}/close-readiness.POST /api/execution-workspaces/{id}/reconcile-branch.
Agent access: partial — REST / native fallback for permitted update; CLI workspace;
reconcile-branch requires board.
Policy: Company/project/workspace visibility and runtime authority; active run, trust preset and task ownership where required.
Evidence: source-only —
implementation at the
reviewed commit above.
Verification: not-run — exercise a listed permitted operation with actual agent
credentials/run context; inspect its result and applicable persistence/attribution.
Assert the relevant denied actor, owner, company or mode from Policy. Review the
remaining operations before qualifying the whole capability.
Gaps: Closure/readiness exists; not every workspace can be reopened, and branch reconciliation has a board boundary.