Files
PaperClipAI/docs/api/secrets.md
T
0b496c9c03 feat(secrets): add run-bound agent secret access (#9921)
## Thinking Path

> - Paperclip is the open source control plane people use to manage
AI-agent companies.
> - Agents already receive selected company secrets through `env.*`
bindings at run launch, but environment injection is ambient,
long-lived, and not suitable for every secret consumer.
> - The existing binding and secret-access-event models already provide
company-scoped authorization and per-resolution audit seams.
> - Agents need an explicit way to discover only the secrets granted to
them and fetch a value on demand without exposing the wider company
catalog.
> - That capability must remain run-bound, preserve low-trust token
carve-outs, and make every value read visible in both security and
operator audit trails.
> - This pull request adds an `access.*` delivery namespace, two
run-bound agent routes, dual audit logging, documentation, and an
operator grants editor.
> - The benefit is least-privilege, revocable, auditable secret access
while preserving existing env injection behavior.

## Linked Issues or Issue Description

No pre-existing public issue. Related work:

- Refs #9797 — existing in-sheet agent access UI that this PR extends to
distinguish env and API delivery.
- Refs #9918 — complementary searchable-agent picker improvement for the
same secrets sheet.
- Refs #9530 — related company-wide metadata catalog proposal; this PR
intentionally exposes only the authenticated run's granted aliases and
values.

**Problem / motivation:** Agents can currently consume secrets only
through process environment injection. This keeps values resident for
the run, does not support on-demand consumers, and cannot provide a
discrete operator-visible activity event for each agent-initiated read.

**Proposed solution:** Treat `company_secret_bindings` as the source of
truth for agent secret grants. Keep `env.KEY` as env delivery and add
`access.ALIAS` for API-only delivery; an env binding also implies read
access because the value is already present in the agent process. Add
run-bound list/fetch endpoints that derive scope from the authenticated
heartbeat run and never accept caller-selected overlays.

**Alternatives considered:** A company-wide agent-readable catalog was
rejected for this value path because it increases reconnaissance and
does not prove a per-secret grant. Reusing the ephemeral
environment-probe resolver was rejected because it lacks binding
enforcement. Approval-gated reads and user-scoped secrets remain
deferred beyond v1.

**Roadmap alignment:** This extends the completed **Secrets Manager with
per-agent access** roadmap capability from launch-time env injection to
explicit run-bound API delivery without duplicating a separate planned
initiative.

## What Changed

- Added `access.*` agent binding validation and a dedicated run-bound
resolver that combines `secrets:read` authorization with binding-context
enforcement.
- Added `GET /api/agents/me/secrets` for minimal granted metadata and
`POST /api/agents/me/secrets/:key/value` for on-demand value fetches
with `Cache-Control: no-store`.
- Preserved the existing denials for low-trust review agents,
task-bridge credentials, and skill-test tokens; standard long-lived
agent API keys cannot call the run-bound routes.
- Added dual audit behavior: value attempts write `secret_access_events`
and `activity_log` (`secret.value.read`), while metadata listing writes
the lighter `secret.access.listed` activity event.
- Kept env compatibility: `env.*` remains injected at launch and also
implies API read for the same bound agent; `access.*` never becomes an
environment variable.
- Added the agent-settings **Secret access** editor plus
delivery-mode/alias surfacing on the Secrets page, with focused UI tests
and tokenized layout styles.
- Updated OpenAPI, shared types, agent-facing skill documentation, and
API reference documentation.

### UI Screenshots

P3 produced and reviewed three screenshots using mock data; images are
intentionally not committed to the repository:

- `secret-access-editor.png` — agent settings grant editor.
- `secret-access-light.png` — Secrets-page delivery surfacing in light
mode.
- `secret-access-dark.png` — Secrets-page delivery surfacing in dark
mode.

The source attachments are retained with the implementation task and
linked in the internal handoff; the public page publisher was
unavailable in the PR-prep runtime.

## Verification

- `pnpm exec vitest run
server/src/__tests__/agent-secrets-routes.test.ts
server/src/__tests__/secrets-service.test.ts
server/src/__tests__/secrets-routes.test.ts
ui/src/lib/secret-delivery.test.ts
ui/src/components/AgentSecretAccessEditor.test.tsx` — 5 files, 122 tests
passed.
- Security follow-up: `pnpm exec vitest run
server/src/__tests__/agent-secrets-routes.test.ts
server/src/__tests__/secrets-service.test.ts` — 2 files, 73 tests passed
after active-run and version-consistency fixes.
- Final-head CI: all feature, typecheck, build, e2e, security, and
review gates pass; `General tests (server (1/3))` remains red after one
rerun because unrelated `heartbeat-retry-scheduling.test.ts` cleanup
deletes `heartbeat_runs` before referenced `activity_log` rows.
- `pnpm --filter @paperclipai/server typecheck` — passed.
- `pnpm --filter @paperclipai/shared typecheck` — passed.
- `pnpm --filter @paperclipai/ui typecheck` — passed.
- `pnpm check:token-gates` — feature-local arbitrary-value violations
fixed; command still reports five unchanged `#9627` literals outside
this PR.
- End-to-end QA passed all eight acceptance criteria: grant/list, fetch,
dual audit, env-implies-read, denial matrix, revocation, UI rendering,
and env-injection regression. Evidence:
https://github.com/paperclipai/paperclip/pull/9921#issuecomment-5027455492
- Security review returned PASS-with-required-changes; the
implementation uses the required dedicated binding-enforcing resolver,
run-bound JWT restriction, run-derived overlays, minimal metadata, and a
resolver redaction-registration hook. Evidence:
https://github.com/paperclipai/paperclip/pull/9921#issuecomment-5027455382

## Risks

- A compromised agent can exfiltrate any secret explicitly granted to
it; explicit company-scoped/run-scoped grants, revocation, and audit
reduce but cannot remove that inherent capability risk.
- The resolver invokes a redaction-registration hook before returning
values, but the current route has no persistent cross-request per-run
redaction registry. Paperclip-owned later comments/events therefore
cannot yet guarantee automatic scrubbing of a deliberately copied
fetched value; QA classified this as non-blocking residual hardening.
- Audit-event insertion currently fails open if the security-event
insert itself fails; the operator activity event provides partial
redundancy, but a future hardening change should define fail-closed
behavior for value delivery.
- This PR overlaps `ui/src/pages/Secrets.tsx` with #9918 and may require
a straightforward rebase after that PR moves.

> 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-5.3-codex`, with reasoning, repository tool use,
terminal execution, Paperclip API access, and GitHub CLI capabilities.
Context-window size is not exposed by the runtime.
- Anthropic Claude Opus 4.8 with 1M context and tool use assisted with
the UI implementation commit.

## 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>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-22 08:04:39 -05:00

15 KiB

title, summary
title summary
Secrets Secrets CRUD

Manage encrypted secrets that agents receive through environment bindings or fetch on demand.

Agent List and Fetch

These routes require the current run-bound agent JWT. They are not available to long-lived agent keys, low-trust review agents, task-bridge keys, or skill-test tokens.

List the secrets accessible to the current run without materializing values:

GET /api/agents/me/secrets
{
  "secrets": [
    {
      "key": "github_token",
      "name": "GitHub token",
      "description": null,
      "delivery": "env",
      "projectionClass": "unclassified",
      "latestVersion": 2,
      "versionSelector": "latest",
      "resolvedVersion": 2
    }
  ]
}

delivery is env, api, or both. The list never returns values, secret IDs, binding IDs, or config paths. An env.* binding implies read access through this API; an access.* binding grants API access without environment injection.

Fetch a value only when it is needed. The request has no body and the response uses Cache-Control: no-store:

POST /api/agents/me/secrets/github_token/value
{
  "key": "github_token",
  "value": "decrypted-secret-value",
  "version": 2
}

Prefer env injection when the adapter or its child processes need the value on every run. Prefer on-demand fetch for values used only on some runs, large or structured values, or skills and tools that do not inherit adapter env. Every successful or failed value fetch is audited in both secret_access_events and activity_log; agents must not log or paste fetched values into issues, comments, or documents.

List Secrets

GET /api/companies/{companyId}/secrets

Returns secret metadata (not decrypted values).

Create Secret

POST /api/companies/{companyId}/secrets
{
  "name": "anthropic-api-key",
  "value": "sk-ant-..."
}

The value is encrypted at rest. Only the secret ID and metadata are returned.

To link a provider-owned secret without copying the value into Paperclip, create an external-reference secret:

{
  "name": "prod-stripe-key",
  "provider": "aws_secrets_manager",
  "managedMode": "external_reference",
  "externalRef": "arn:aws:secretsmanager:us-east-1:123456789012:secret:paperclip/prod/stripe",
  "providerVersionRef": "version-id-or-label"
}

Paperclip stores the provider reference and a non-sensitive fingerprint only. The value is resolved, when the provider is configured, through the server runtime path that enforces binding context and records access events.

Provider Health

GET /api/companies/{companyId}/secret-providers/health

Returns provider setup diagnostics, warnings, and local backup guidance. Health responses must not include secret values or provider credentials.

For aws_secrets_manager, an unready health response names the missing non-secret provider environment variables, the AWS SDK default credential source expected by the server runtime, and the custody rule that AWS bootstrap credentials must not be stored in Paperclip company_secrets.

The equivalent CLI check is:

pnpm paperclipai secrets doctor --company-id {companyId}

Provider Vaults

Provider vaults are named, company-scoped configurations that route secret material to one of the supported provider backends. See the secrets deploy guide for the operator model and custody rules.

All routes below require board auth and company access. Mutating routes emit secret_provider_config.* activity-log entries. No route in this surface returns provider credential values; submitting credential-shaped fields in config is rejected at validation time.

List Vaults

GET /api/companies/{companyId}/secret-provider-configs

Returns every vault for the company (including disabled rows for audit), each with id, provider, displayName, status, isDefault, non-sensitive config, latest health snapshot (healthStatus, healthCheckedAt, healthMessage, healthDetails), disabledAt, and audit columns.

Create Vault

POST /api/companies/{companyId}/secret-provider-configs
{
  "provider": "aws_secrets_manager",
  "displayName": "Prod US-East",
  "isDefault": true,
  "config": {
    "region": "us-east-1",
    "namespace": "paperclip",
    "secretNamePrefix": "paperclip",
    "kmsKeyId": "arn:aws:kms:us-east-1:123456789012:key/abcd-...",
    "environmentTag": "production"
  }
}

Per-provider config shapes:

  • local_encrypted: optional backupReminderAcknowledged: boolean.
  • aws_secrets_manager: required region; optional namespace, secretNamePrefix, kmsKeyId, ownerTag, environmentTag.
  • gcp_secret_manager (coming soon): optional projectId, location, namespace, secretNamePrefix.
  • vault (coming soon): optional origin-only HTTPS address, namespace, mountPath, secretPathPrefix. address values with embedded credentials, paths, query strings, or fragments are rejected.

status defaults to ready for local_encrypted and aws_secrets_manager, and to coming_soon for gcp_secret_manager and vault. Coming-soon and disabled vaults cannot be marked isDefault. Setting isDefault: true clears the previous default for the same provider in the same transaction.

Get Vault

GET /api/secret-provider-configs/{id}

Update Vault

PATCH /api/secret-provider-configs/{id}
{
  "displayName": "Prod US-East-2",
  "config": {
    "region": "us-east-2",
    "kmsKeyId": "arn:aws:kms:us-east-2:123456789012:key/abcd-..."
  }
}

config is replaced wholesale on update — pass the full provider config payload, not a partial diff. Status transitions for gcp_secret_manager and vault are constrained to coming_soon and disabled until their runtime modules ship.

Disable Vault

DELETE /api/secret-provider-configs/{id}

Soft-deletes the vault: status flips to disabled, isDefault clears, and disabledAt is stamped. Disabled vaults remain in GET results for audit purposes but are no longer offered in the secret create/rotate flow.

Set Default

POST /api/secret-provider-configs/{id}/default

Marks the target vault as the default for its provider family and clears the previous default. Returns 422 when the target is coming_soon or disabled.

Run Health Check

POST /api/secret-provider-configs/{id}/health

Runs a provider-specific health probe and persists the result on the vault. Response shape:

{
  "configId": "<uuid>",
  "provider": "aws_secrets_manager",
  "status": "ready" | "warning" | "error" | "coming_soon" | "disabled",
  "message": "Provider vault is ready to handle managed writes",
  "details": {
    "code": "provider_ready",
    "message": "...",
    "guidance": ["..."]
  },
  "checkedAt": "2026-05-06T14:00:00.000Z"
}

Health responses never include provider credentials or secret values. For AWS vaults, details.guidance may include missing non-secret env names and the expected AWS SDK credential source; coming-soon vaults always return status: "coming_soon" with code: "runtime_locked" and never call into provider modules.

Selecting A Vault When Creating Or Rotating Secrets

POST /api/companies/{companyId}/secrets and POST /api/secrets/{secretId}/rotate both accept an optional providerConfigId field that pins the secret to a specific vault. When omitted (or null), the operation runs through the deployment-level provider configuration — the same path existing installs already use. The board UI preselects the company's default vault for the chosen provider before submitting, so callers should usually send an explicit providerConfigId. Coming-soon and disabled vaults are rejected with a 422; a vault that does not match the secret's provider is rejected the same way.

POST /api/companies/{companyId}/secrets
{
  "name": "prod-stripe-key",
  "provider": "aws_secrets_manager",
  "providerConfigId": "<vault-uuid>",
  "managedMode": "external_reference",
  "externalRef": "arn:aws:secretsmanager:us-east-1:123456789012:secret:paperclip/prod/stripe"
}

Response Redaction Rules

Every route in this surface enforces the same redaction contract:

  • Secret values are never returned. The board UI never has a "reveal value" affordance; resolution happens server-side at runtime under a binding.
  • Provider credential values are never accepted, stored, returned, logged, or echoed in error messages. Submitting credential-shaped fields fails validation with a non-leaking error.
  • Activity log entries record vault id, provider, displayName, status, and isDefault transitions — never config payloads or health detail bodies.

Remote Import From AWS Secrets Manager

Remote import links existing AWS Secrets Manager entries into Paperclip as external_reference secrets. Import stores provider reference metadata only; it does not copy the remote secret plaintext into Paperclip.

The routes are board-only and company-scoped. providerConfigId must point to a same-company AWS provider vault with status ready or warning. Disabled, coming-soon, non-AWS, and cross-company vaults are rejected. Imported secrets resolve later through the selected vault, so runtime reads still need secretsmanager:GetSecretValue and any required KMS decrypt permission on the selected external secret.

Preview Remote Import Candidates

POST /api/companies/{companyId}/secrets/remote-import/preview
{
  "providerConfigId": "<aws-vault-uuid>",
  "query": "stripe",
  "nextToken": "opaque-provider-token",
  "pageSize": 50
}

query is optional and is passed to AWS Secrets Manager inventory filtering. Treat it as non-secret metadata because AWS may record list request parameters in CloudTrail. nextToken is an opaque AWS cursor; callers must pass it back unchanged and must not synthesize offsets. pageSize is optional, defaults to 50 in the UI, and is capped at 100.

Preview uses AWS ListSecrets only. It must not call GetSecretValue or BatchGetSecretValue, must not request SecretString, and must not require KMS decrypt. The response contains sanitized metadata for display and conflict decisions:

{
  "providerConfigId": "<aws-vault-uuid>",
  "provider": "aws_secrets_manager",
  "nextToken": null,
  "candidates": [
    {
      "externalRef": "arn:aws:secretsmanager:us-east-1:123456789012:secret:prod/stripe",
      "remoteName": "prod/stripe",
      "name": "prod/stripe",
      "key": "prod-stripe",
      "providerVersionRef": null,
      "providerMetadata": {
        "createdDate": "2026-05-06T00:00:00.000Z",
        "lastChangedDate": "2026-05-06T00:00:00.000Z",
        "hasDescription": true,
        "hasKmsKey": true,
        "tagCount": 3
      },
      "status": "ready",
      "importable": true,
      "conflicts": []
    }
  ]
}

Candidate statuses:

  • ready: the row can be selected for import.
  • duplicate: a Paperclip secret already links the same canonical provider reference for the same provider vault.
  • conflict: the row has a name/key collision or provider guardrail failure.

Conflict types are exact_reference, name, key, and provider_guardrail. AWS refs under Paperclip's own managed namespace are blocked as external references; use the Paperclip-managed secret flow for those resources instead.

Import Selected Remote References

POST /api/companies/{companyId}/secrets/remote-import
{
  "providerConfigId": "<aws-vault-uuid>",
  "secrets": [
    {
      "externalRef": "arn:aws:secretsmanager:us-east-1:123456789012:secret:prod/stripe",
      "name": "Stripe production key",
      "key": "stripe-production-key",
      "description": "Stripe key used by production checkout",
      "providerVersionRef": null,
      "providerMetadata": {
        "createdDate": "2026-05-06T00:00:00.000Z"
      }
    }
  ]
}

The secrets array accepts 1-100 rows. Each row may override the suggested Paperclip name, key, optional Paperclip description, providerVersionRef, and sanitized providerMetadata. Blank descriptions are stored as null; AWS provider descriptions are not copied into Paperclip descriptions. The backend re-checks duplicate refs and name/key conflicts at submit time; a stale preview does not bypass those checks.

The import response is row-level:

{
  "providerConfigId": "<aws-vault-uuid>",
  "provider": "aws_secrets_manager",
  "importedCount": 1,
  "skippedCount": 1,
  "errorCount": 0,
  "results": [
    {
      "externalRef": "arn:aws:secretsmanager:us-east-1:123456789012:secret:prod/stripe",
      "name": "Stripe production key",
      "key": "stripe-production-key",
      "status": "imported",
      "reason": null,
      "secretId": "<paperclip-secret-id>",
      "conflicts": []
    }
  ]
}

Row statuses:

  • imported: Paperclip created an active external_reference secret and one metadata-only version row.
  • skipped: the row had an exact-reference duplicate or name/key conflict.
  • error: the provider rejected the reference or the row failed validation.

Activity logs for preview/import store aggregate counts, provider id, and vault id only. They must not store remote secret names, ARNs, descriptions, tags, plaintext values, provider credentials, or raw AWS error blobs.

Rotate Secret

POST /api/secrets/{secretId}/rotate
{
  "value": "sk-ant-new-value..."
}

Creates a new version of the secret. Agents referencing "version": "latest" automatically get the new value on next heartbeat. Pin to a specific version when a bad latest rollout would affect many agents at once.

Using Secrets in Agent Config

Reference secrets in agent adapter config instead of inline values:

{
  "env": {
    "ANTHROPIC_API_KEY": {
      "type": "secret_ref",
      "secretId": "{secretId}",
      "version": "latest"
    }
  }
}

The server resolves and decrypts secret references at runtime, injecting the real value into the agent process environment. Paperclip's custody guarantees end at injection: the agent process can read, log, or forward the value, so treat any secret bound to an agent as exposed to that agent. See the custody boundaries note in the secrets deploy guide.

User-specific env bindings use a definition key instead of a concrete secretId. The concrete value is resolved for the run's responsible user:

{
  "env": {
    "GITHUB_TOKEN": {
      "type": "user_secret_ref",
      "key": "github_api_token",
      "version": "latest",
      "required": true,
      "allowMissingOverride": false
    }
  }
}

required defaults to true and allowMissingOverride defaults to false. Missing required user-secret values must fail closed before adapter dispatch. Optional missing values omit the environment variable; they must not inject an empty string or another user's value. Paperclip records value-free access events with secretScope, responsibleUserId, credentialOwnerUserId, and userSecretDefinitionId.

Portability

Company export/import APIs represent agent and project environment requirements as declarations in the package manifest. Exports omit secret values, secret IDs, provider references, and encrypted provider material. Use:

pnpm paperclipai secrets declarations --company-id {companyId}

to inspect the declarations that an export would emit before moving a package.