Files
PaperClipAI/doc/skill-ephemeral-web-servers.md
T
DottaandPaperclip 0e15c92073 feat(skills): make managed content independent of web-server disk (#18334)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work.
> - Agents use installed skills and exact version pins during execution.
> - Managed skills still depended on mutable directories on the web
server.
> - A replacement server could lose acknowledged edits or remove an
unassigned skill whose directory was absent.
> - This pull request makes immutable versions the authority for managed
content and installed repository reads.
> - The server can rebuild delivery files from durable content after
replacement.
> - This establishes skills readiness for ephemeral web servers. General
server draining and workspace recovery remain separate work.

## Linked Issues or Issue Description

**Subsystem affected**

Skills services, shared contracts, database receipts, Skill Studio, and
runtime delivery.

**Problem or motivation**

A fresh web server must discover, read, edit, and deliver protected
installed skills using the same database, object storage, secrets, and
supported application artifacts. It must not need its predecessor's
files or process memory.

**Proposed solution**

Use `managed` ownership with a null source locator. Read installed
content by immutable version. Prepare packages before a short
publication transaction. Persist execution selections before
materialization. Recover exact pinned runtime assets and reject deletion
while active or recovering pins exist.

**Alternatives considered**

Reconstructing a mutable installed directory preserves the old
source-of-truth ambiguity. Permanent dual writes create two competing
authorities. This change uses one managed model after a bounded legacy
conversion.

**Roadmap alignment**

This improves the existing Skills Manager, Studio, version history, and
Cloud execution capabilities. The maintainer supplied the implementation
contract. This does not add a separate skills product or deploy fleet
changes.

Related work: #18082 supplies immutable package storage. #17273
addresses edits outside the API for linked local skills; this PR keeps
that source category separate. #2648 is a closed earlier
ephemeral-filesystem proposal.

## What Changed

- Add managed ownership and focused content publication and execution
materialization modules. Keep the archive format and package storage
tables.
- Publish managed create, save, delete, fork, rename, restore, and
catalog content with version checks. Add durable create receipts.
Preserve file-save receipt responses.
- Return file version identities. Keep Studio drafts after a conflict
and provide an explicit reload action. A delayed reload cannot overwrite
newer typing or navigation.
- Persist exact runtime selections, including empty selections. Rebuild
pinned assets and verify their admitted digests on a replacement server.
- Preserve source access on forks and disconnected copies. Keep the
existing workspace policy category for managed skills.
- Add bounded conversion, eligibility, and independent verification
phases. Load persisted Cloud configuration in the standalone CLI without
seeding or changing it. Reject partial current versions against
installed inventories; recover only complete saved content or the exact
source. Document storage migration blockers and the compatible rollback
floor.
- Add separate-process PostgreSQL/S3 qualification and update domain,
retention, policy, and UI regressions.
- Pass the plugin worker manager into Studio execution, admit restricted
`skill_test` mode through native Runner contracts, and preserve admitted
scope across native and REST tools. Native tests cannot create ordinary
tasks, browse company data, or use external tools.
- Select the exact durable Studio revision even when unassigned or Beta
skills is off. Persist Studio output and audit atomically with native
task completion, preserving the result on replay.

## Verification

- Final revision `13e8af480b88ba6fea5737bac4bd81104d604792` passed full
CI: all 49 workflow jobs completed successfully ([run
41127396098](https://github.com/paperclipai/paperclip/actions/runs/41127396098)).
This includes typecheck, test suites, Runner checks, browser checks and
build. Greptile review 15 is 5/5 on that same revision, with zero
unresolved threads. The PR is ready for review.
- Local workspace typecheck and build passed after integrating master.
Focused checks passed: 33 execution/completion cases, 52 Runner HTTP
cases, 16 managed-content cases, 17 Studio cases, the new real-process
CLI regression, map checks and UI token gates. The CLI regression proves
inspection does not seed missing Cloud configuration. Server typecheck
passed after that fix.
- The independent-process PostgreSQL/S3 matrix passed through normal
HTTP skill routes. It covers A → B → C replacement, lost commit
responses, upload crashes, concurrent saves, binary contents and
executable delivery. Metadata listing and warm selected delivery retain
zero object downloads. Local timing measurements are recorded; they do
not establish hosted latency.
- Hosted A → B → C passed on the owned disposable test tenant with
external Postgres, S3 and a real remote Paperclip Computer. Predecessor
temporary homes were absent. C read and executed B's acknowledged
managed save. New runs used repository v1 while v2 was installed. Hashes
matched an independently checked exact upstream commit.
- Compatible image rollback to `6518b323` passed. Studio retained the
unassigned fork's C save. A fresh remote run executed the newer managed
content and historical repository pin. The actual shell event confirmed
expected hashes, stdout, exit 0 and mode 0555.
- Runtime candidate `9200e3fcb4` deployed successfully ([Cloud run
41124147876](https://github.com/paperclipai/paperclip-cloud/actions/runs/41124147876)).
The final Studio test delivered only the unassigned fork's exact v3,
executed its script and persisted a successful output snapshot.
Independent command-event inspection confirmed all three hashes, stdout,
exit 0 and mode 0555. All 14 installations were version-backed.
Independent verification checked all 25 versions: 23 remote objects, two
bundled database snapshots, zero blockers.
- The deployed image is `9200e3fcb4`. Later master integration, the
client-only delayed-reload draft guard and the CLI bootstrap have
separate automated proof. The CLI bootstrap was also exercised against
the tenant's stored configuration and storage through the same read-only
services. This PR does not claim that the deployed image is the final PR
revision.
- Live Studio checks exposed worker wiring, admission, access scope,
exact selection and completion-publication gaps. They are fixed. Failed
attempts remain in the evidence. The full local suite exceeded 80
minutes under host contention and was stopped. Later combined/isolated
local attempts also timed out; they are not counted as passes. Full CI
covers the integrated suites. Earlier CI found an obsolete restart-scope
fixture, which was corrected and passed all 52 HTTP cases.
- Reproduction and migration:
[`doc/skill-ephemeral-web-servers.md`](doc/skill-ephemeral-web-servers.md).
Source revisions, deployments, versions, run IDs, hashes and failure
history:
[`doc/plans/2026-10-30-ephemeral-web-skills.md`](doc/plans/2026-10-30-ephemeral-web-skills.md).

## Risks

- Legacy conversion requires stopped admission and retired
filesystem-authoritative writers. Pre-conversion binaries are not
supported rollback targets afterward.
- Cloud readiness requires remote object storage. Local provider
receipts and provider mismatches remain blockers until objects are
copied and verified.
- Linked local sources retain filesystem semantics and are outside the
cloud guarantee.
- Cold reads still fetch the existing archive format. Warm runtime
preparation retains the verified-cache path.
- Bundled reconciliation preserves an installed head. Existing release
snapshots remain selectable. An older server must not replace a newer
head with its shipped files.
- No production volumes, fleet defaults, or general deployment
coordination change in this PR.

## Model Used

OpenAI Codex, GPT-6 family, with reasoning, repository tools, code
execution, and browser use. The exact backend model ID and
context-window size are not exposed in this session. No subagents were
used.

## 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>
2026-10-30 16:18:27 -05:00

10 KiB

Skills on ephemeral web servers

The skills guarantee is replacement of A by B, and then C, with empty web-server homes and the same database, object storage, secrets and supported application artifacts. Disk remains useful for scratch archives, verified caches and transport staging. It does not own acknowledged managed edits or installed version selection. This guarantee does not qualify instructions, workspaces, controller state, run logs, traffic switching or readiness/draining for volume removal.

Ownership and publication

managed is Paperclip ownership. Its locator is null; editability does not depend on a directory's position beneath a managed root. Existing policies normalize it to workspace, the policy category previously used by managed local skills. Repository provenance and tracking remain separate from installed content. Forks and disconnected editable copies retain source access and cannot publish restricted content through a public source category.

managed-skill-content.ts reads immutable version inventories, prepares candidates, and publishes them against a captured head. Package upload and verification happen before the final database transaction. That transaction rechecks identity/access, locks names in sorted order, locks the skill, verifies the base, and publishes the version, head, projections and applicable mutation audit/receipt. Publication races return 409; clients must reconcile instead of automatically rebasing. Complete inline versions remain the compatibility fallback when external writes are disabled; they never depend on an installed directory. Normal cloud qualification uses remote object storage.

The initial create response, including its original version, can be replayed after a lost response and subsequent edits. Its durable API receipt retains the response projection without Markdown; the immutable version supplies the body. Receipts remain scoped to company, actor and operation context. A deleted skill's receipt fails explicitly rather than creating another identity.

skill-execution-content.ts owns temporary materialization. Heartbeat preparation selects authorized immutable versions, saves the selection (including an empty selection), and only then materializes admitted entries while retaining the existing retention lock. A renamed skill is recovered through the version's stable skill ID. Native recovery resolves the version, reconstructs its v1 projection and requires the admitted digest, manifest digest, file count and byte count to match. Adjacent compatible releases must preserve that projection contract.

Studio tests deliver the selected saved version through the same native Runner path, even without a permanent agent assignment. Their skill_test mode remains explicit in the execution input. The durable test row replaces ordinary agent assignments with its exact version, independently of the Beta skills preference; required environment skills remain subject to their existing policy. A shadowed or inaccessible test skill fails preparation. Preconfigured Runner homes cannot receive this managed delivery, so Studio requires an isolated home. Task-bound credentials and native authorization retain the admitted scope through restart and task-mode edits. Native tools expose only the harness task and admitted skills; company browsing, task creation, connectors and assigned MCP tools are unavailable. A rollback image must understand that mode to continue an active Studio test. Settle tests before rolling back to an earlier image that rejects it.

Native task completion publishes the Studio result and output snapshot in the same transaction as its status decision. Replaying a committed decision cannot recapture later document edits or duplicate completion activity. A successful provider turn alone does not complete a test whose task remains unfinished.

Bundled release snapshots remain immutable. Reconciliation preserves an installed head; an older process must never promote its shipped bytes over another installed version. The shipped release history remains independently selectable. Linked-local skills retain mutable source behavior and are explicitly excluded from cloud readiness.

One-time conversion

First deploy compatible readers and shared/API types. Preserve the old volume and keep legacy writers until the final supported contents can be captured. Automatic JSONB package migration is separate from installation conversion; passing one does not prove the other is finished.

Run the existing bounded CLI with the target's explicit database and storage config:

pnpm --filter @paperclipai/server exec tsx scripts/migrate-skill-packages.ts --phase=eligibility --company=<uuid> --limit=10
pnpm --filter @paperclipai/server exec tsx scripts/migrate-skill-packages.ts --phase=installations --company=<uuid> --limit=10

These commands return a cursor. Continue each phase with --after=<uuid> until an empty batch. Eligibility is a metadata-only report, not a storage verification claim. Installation preview inspects candidate legacy contents without publishing them. For remote legacy imports it can acquire their exact pinned source; it must never use the latest branch as a replacement for missing installed content.

  1. Inventory all companies and source kinds. Identify missing versions, incomplete inventory, local storage receipts, catalog reset snapshots and unresolved runs.

  2. Backfill legacy package/version bodies with the existing snapshots and versions phases. The latter also fills missing version metadata indexes so subsequent library/history reads need no object downloads. Keep original bytes until independent verification succeeds.

  3. Stop admission and skill/source writers; retire every filesystem-authoritative process. The --writers-stopped flag attests this operator action; it does not stop the application itself.

  4. Apply final installation conversion on the host with the legacy files:

    pnpm --filter @paperclipai/server exec tsx scripts/migrate-skill-packages.ts --phase=installations --company=<uuid> --limit=10 --apply --writers-stopped
    
  5. Sweep eligibility and verification, including all historical versions. The latter verifies object bytes independently of caches. Follow all cursors; a successful first batch is not a complete readiness report.

  6. Start a compatible server with a fresh home and test list, file read, authoring, history and exact selected execution before reopening normal admission.

Managed conversion preserves skill IDs, keys, history, assignments, pins and access provenance. Identical complete snapshots reuse their version IDs. Catalog conversion preserves the original reset version as well as installed content. A missing legacy directory can use a complete verified saved version. A partial Markdown-only rescue, unknown revision, unsafe source tree or active legacy selection without exact version identity is a blocker. Settle or snapshot such runs while A is still available; do not cancel them simply to pass preflight.

Report statuses distinguish linked-local sources, legacy managed/catalog records, missing current versions, unknown active-run selections, missing/corrupt packages, local storage and provider mismatch. Expired unused Git history keeps its existing 30-day/exact-commit policy. Current, pinned, authored and recovering versions need independent durable content. local_storage is supported self-hosted durability; it does not satisfy remote empty-disk recovery.

Storage and rollback

Cloud readiness requires external Postgres and remote object storage. Migrate and verify existing local-disk objects before changing provider/bucket configuration; changing configuration alone does not copy receipts or objects. Keep the old provider available until a provider migration has copied and independently verified every referenced object and atomically updated its storage receipt. This command does not perform provider migration: local_storage and provider_mismatch are blocking results for cloud qualification, not instructions to switch configuration and retry. A directory of extracted caches is never a durable object backend.

Conversion establishes the rollback floor. Afterward only compatible version-reading binaries may overlap or be rolled back against the same remote state. Reattaching an old volume does not roll back committed content. Filesystem-authoritative binaries require an explicit reverse migration and are unsupported rollback targets. A routine compatible A → B upgrade needs no skills conversion or filesystem copy.

A current version ID alone does not establish conversion completeness. Eligibility and independent verification compare its file inventory with the installed record and report incomplete_current_version when supporting files are missing. Installation conversion can reacquire a complete exact pinned source; unavailable or unpinned sources remain blockers. Historical versions are not compared with newer installed inventories, since later revisions may legitimately add files.

Qualification

The focused service suite covers publication rollback/conflict, supporting and binary files, executable modes, restore/fork/rename, active-run deletion protection and legacy conversion. The opt-in process suite starts independent Node servers with one database connection each, exercises real HTTP routes and object storage, removes predecessor homes and executes a delivered harmless script. It also injects a lost response, failed upload and process death during upload. It is a skills-router qualification, not a full hosted Runner or whole-application deployment claim.

Use owned disposable Postgres and S3-compatible fixtures. The optional admin URL creates a random database per test suite and never migrates the database named in that URL. The S3 test creates and removes its own random bucket. Supply test-only AWS credentials through the environment:

PAPERCLIP_TEST_POSTGRES_ADMIN_URL=<disposable-postgres-url> \
PAPERCLIP_TEST_S3_ENDPOINT=<disposable-s3-endpoint> \
pnpm exec vitest run server/src/__tests__/ephemeral-skills-process.test.ts

Record revision, fixture image digests, version IDs, cache/object reads, cold/warm timings and useful output. These local timings do not establish hosted latency. Separately qualify one disposable staging tenant and a real remote Runner, including an explicit repository import, historical pin, fresh-home replacements, compatible rollback and recovery without predecessor paths. Publish the observed evidence with the PR; a passing local suite is not that staging evidence.