Dotta
dbebf30c89
## Thinking Path > - Paperclip is a control plane for AI-agent companies, so execution policy and trust boundaries are part of the product's safety contract. > - Low-trust review work needs narrower authority than normal same-company agents because hostile PRs, comments, attachments, and generated output can carry prompt-injection payloads. > - The current V1 shape gives trusted workers broad company context, which is useful for normal execution but too permissive for a reviewer assigned to hostile content. > - This branch adds a `low_trust_review` preset, source-trust tagging, route-level containment, and quarantine handling so low-trust output does not automatically flow into higher-trust wake context. > - The branch has been rebased onto current `origin/master`, and the low-trust migration was renumbered to `0097_low_trust_source_trust.sql` to avoid collisions with existing `0091` through `0096` migrations. > - Greptile feedback was addressed by tightening low-trust detection, preserving project-level trust policy checks, fixing issue-kind promotion lookup, removing duplicate post-lease isolation assertion, documenting fail-closed source-trust behavior, bounding ancestry checks, enforcing runtime issue context for CEOs, awaiting accepted-plan monitor authorization, and making low-trust issue source-trust tagging atomic. > - The benefit is a first production slice of deny-by-default review containment with regression coverage for the main control-plane pivot surfaces. Fixes #7531. ## What Changed - Added shared trust-policy types and validators, plus database/source-trust fields for issues, comments, documents, and work products. - Implemented server enforcement for low-trust issue scope, agent self-view redaction, secret/plugin/runtime denial paths, promotion checks, and quarantined continuation/wake context. - Added focused low-trust regression tests for resolver behavior, source trust, route authorization, heartbeat preflight ordering, runtime containment, and quarantine redaction. - Added board UI affordances for selecting/reviewing the low-trust preset and surfacing source-trust badges in relevant issue views. - Added `doc/LOW-TRUST-PRESETS.md`, updated `doc/SPEC-implementation.md`, and committed the low-trust review contract plan under `doc/plans/`. - Rebasing note: the original `0097_low_trust_source_trust.sql` migration was renamed to `0097_low_trust_source_trust.sql`; the SQL uses `ADD COLUMN IF NOT EXISTS` so users who already applied the old-numbered migration are not broken by the renumbered migration. ## Verification - Rebased branch onto current `origin/master` and force-pushed with lease to `origin/PAP-10211-low-trust-agent` at head `2719f31e3`. - Confirmed the PR diff does not include `pnpm-lock.yaml` or `.github/workflows` changes. - Resolved upstream UI/comment conflicts by preserving deleted-comment tombstone behavior and low-trust source-trust badges/metadata. - Renumbered the low-trust source-trust migration to `0097_low_trust_source_trust.sql`; the SQL uses `ADD COLUMN IF NOT EXISTS` so users who already applied an old-numbered copy are not broken. - `pnpm exec vitest run ui/src/lib/issue-chat-messages.test.ts server/src/__tests__/heartbeat-workspace-session.test.ts` - `pnpm exec vitest run server/src/__tests__/source-trust.test.ts server/src/__tests__/workspace-runtime-service-authz.test.ts ui/src/lib/trust-policy-ui.test.ts ui/src/components/TrustPresetSection.test.tsx` - `pnpm run typecheck:build-gaps` - `git diff --check` - GitHub checks pass on head `2719f31e3`: build, typecheck/release registry, general tests, serialized server suites, e2e, canary, verify, policy/review, Socket, and Snyk. - Greptile Review passes with Confidence Score 5/5 and zero unresolved Greptile review threads. - No design screenshots/images were added because the task explicitly says not to add them unless they are specifically part of the work. ## Risks - Medium risk: this touches shared trust-policy contracts, server authorization paths, heartbeat context generation, migration metadata, and UI preset controls. - Low-trust containment is intentionally deny-by-default; legitimate future review workflows may need explicit allowlisted exceptions. - Plugin/runtime/security surfaces are broad, so regression tests cover the current known routes but future integrations must route through the same containment layer. - The PR is ready for review; GitHub checks are green and Greptile is 5/5. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - OpenAI Codex, GPT-5 coding agent, tool-enabled shell and GitHub CLI workflow. ## 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 run tests locally and they pass - [x] I have added or updated tests where applicable - [x] UI changes are covered by focused tests; no screenshots were added per task instruction not to add design images unless specifically required - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>
56 lines
2.4 KiB
Markdown
56 lines
2.4 KiB
Markdown
# Low-Trust Presets
|
|
|
|
Paperclip ships core trust preset names so containment decisions are enforced in
|
|
Community Edition even when EE policy editing is unavailable.
|
|
|
|
## Presets
|
|
|
|
- `standard`: the default V1 company-visible collaboration model. This preserves
|
|
existing behavior for normal agents.
|
|
- `low_trust_review`: an opt-in containment preset for automated work that may
|
|
consume hostile or prompt-injected input, such as untrusted pull requests,
|
|
external tickets, dependency diffs, or generated review output.
|
|
|
|
## Boundary Model
|
|
|
|
`low_trust_review` is resolved from existing JSON policy fields:
|
|
|
|
- agent permissions: `permissions.trustPreset` and
|
|
`permissions.authorizationPolicy.trustBoundary`
|
|
- project policy:
|
|
`executionWorkspacePolicy.authorizationPolicy.trustBoundary`
|
|
- issue/run policy: `executionPolicy.authorizationPolicy.trustBoundary`
|
|
|
|
The resolver intersects those sources. Narrower wins. A low-trust preset must
|
|
resolve to a concrete company-local project, root issue, or issue-id scope. If a
|
|
policy source names another company, uses an unsupported preset, or lacks that
|
|
scope for risky access, Paperclip fails closed.
|
|
|
|
## Containment, Not Privacy
|
|
|
|
This is containment for hostile automated work. It is not a general project,
|
|
issue, or human privacy system.
|
|
|
|
V1 standard work remains company-visible by default: board users and in-company
|
|
actors can inspect company work objects unless a separate access-control feature
|
|
changes that behavior. Low-trust containment instead limits what the low-trust
|
|
agent can read or mutate through the Paperclip API and prevents raw untrusted
|
|
output from being automatically promoted into higher-trust agent context.
|
|
|
|
## Runtime Containment
|
|
|
|
Managed `low_trust_review` runs fail closed unless Paperclip can enforce the
|
|
runtime boundary:
|
|
|
|
- the selected execution environment must use the `sandbox` driver
|
|
- the effective execution workspace mode must be `isolated_workspace`
|
|
- the issue being run must be inside the resolved low-trust boundary
|
|
- secret references must use binding ids explicitly allowed by the boundary
|
|
- inline sensitive environment values such as API keys and tokens are rejected
|
|
- workspace runtime-service mutations are denied unless the boundary explicitly
|
|
grants the `runtime.manage` tool class
|
|
|
|
The Docker workflow in `doc/UNTRUSTED-PR-REVIEW.md` remains useful for manual
|
|
local review, but Paperclip-managed low-trust execution requires a sandboxed
|
|
environment instead of a host-local adapter process.
|