The mutex already serializes all writes, so the PID-based temp name
is unnecessary. A fixed suffix matches writeStore and avoids orphaned
temp files after a crash (a restarted process picks a new PID).
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Add integration tests that directly exercise the PAPERCLIP_API_URL
conditional guard in startServer():
- Externally set PAPERCLIP_API_URL is preserved (not overwritten)
- Unset PAPERCLIP_API_URL falls back to host-based URL
Addresses Greptile review comment on PR #3472.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Add a `dryRun` option to the scan-projects endpoint. When true, the
scan identifies which skills would be pruned and which agents would be
affected, but does not delete anything or modify agent configs.
The response now includes:
- `pruned[]`: list of skills that would be (or were) removed, with
affected agent names
- `dryRun`: boolean echoed back so callers can distinguish preview
results from live mutations
This lets callers preview destructive prune operations before committing
to them, addressing the review concern about silent deletion of
production data.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Previously, skills pruned during re-scan only emitted a warning when
they were attached to agents. Skills with no agent references were
deleted silently. Now every pruned skill emits a warning in the scan
result so the deletion is always visible to the caller.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Cover the pruning logic in scanProjectWorkspaces with integration tests
using embedded Postgres and mocked GitHub HTTP layer:
- skill removed from source → detached from agents → deleted from DB
- skill removed with no agent references → deleted without warnings
- source fetch failure → no pruning, warning emitted
Addresses review feedback on PR #3351 requesting test coverage for the
destructive prune/delete behavior.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
During a repo sync (scan button), if a skill was removed from the source
and is still assigned to agents, the skill is now automatically detached
from those agents before deletion, rather than leaving it attached.
Manual delete-by-source is unchanged (still blocks if in use).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
GitHub/sks_sh pruned skills don't have project/workspace context needed for
the CompanySkillProjectScanConflict/Skipped types. Orphaned skills are
silently deleted; skills still used by agents emit a warning instead of
a conflict entry.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Verifies the scan endpoint returns skipped and conflicts arrays in the
response, covering the prune path where removed skills are either
skipped (orphaned) or added to conflicts (still used by agents).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When re-scanning existing sources, diff skills in the DB against the
current source manifest. Skills no longer in the source are either:
- Added to conflicts (if still used by agents)
- Deleted via deleteSkill (if orphaned), added to skipped
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Verifies that agents with canCreateAgents permission can call the
scan-projects endpoint and that scanProjectWorkspaces is invoked.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Use Set<string> instead of Map for tracking unique source locators
- Remove dead skillsAtSource destructuring from loop
- Remove redundant deriveCanonicalSkillKey call (already set by readUrlSkillImports)
- Invert slug check to continue guard for cleaner flow
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adapters installed with the old --no-save flag are not tracked in
package.json and get pruned when another adapter is installed. This
adds ensurePackageOnDisk() which detects missing packages and
reinstalls them from npm before reload or server boot attempts to
import them, fixing ENOENT errors for previously-pruned adapters.
Fixes FAR-47
Co-Authored-By: Paperclip <noreply@paperclip.ing>
npm install --no-save does not record dependencies in package.json,
so when a second adapter is installed, npm prunes the first as
extraneous. Removing --no-save ensures all installed adapters are
tracked in the managed package.json and persist across installs.
Fixes FAR-47
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Add an async mutex (promise-chain FIFO queue) to serialise all adapter
mutation routes (install, reinstall, reload, delete) so that concurrent
requests cannot race on npm, the adapter-plugins.json store, or the
in-memory adapter registry.
Also switch adapter-plugin-store file writes to atomic write-tmp-then-rename
to prevent partial/corrupted reads from concurrent processes.
Includes the packageName.trim() fix for whitespace-induced npm failures.
9 new tests covering mutex serialisation, error recovery, FIFO ordering,
and atomic store operations.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Previously the server unconditionally overwrote PAPERCLIP_API_URL on
startup, clobbering any value set externally (e.g., via Kubernetes
ConfigMap). Now only sets the default if the env var is not already set.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Replace redundant listFull() call with acceptedSkills to avoid extra DB round-trip
- Check slug conflicts against full acceptedSkills list instead of just same-source skills
- Call upsertAcceptedSkill after persisting to keep in-memory list current
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When "Scan project workspaces for skills" runs, now also iterates all
existing GitHub/sks_sh skills and re-fetches their source repos to
detect newly added skills. New skills are upserted automatically.
Skips sources that fail, logged as warnings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When updateSkillAuth(null) is called, the underlying secret row was
left orphaned. Now deletes the secret via secretsSvc.remove() before
clearing sourceAuthSecretId from metadata.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Scope metadata update WHERE clause to companyId for defence-in-depth
- Add CompanySkillUpdateAuth inferred type export to match other schemas
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Pre-check all skills for agent usage before deleting any in deleteBySource
to prevent partial/failed deletions
- Delete (rotate to empty) the skill-pat:<skillId> secret when a skill is
deleted to prevent orphaned PAT secrets
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add optional authToken to skill import for GitHub private repos
- Store PAT as encrypted company secret (skill-pat:{skillId})
- Thread auth token through ghFetch, fetchText, fetchJson, and all GitHub resolution functions
- Add PATCH /companies/:companyId/skills/:skillId/auth for managing PAT per skill
- Add DELETE /companies/:companyId/skills/by-source for bulk deleting skills from a repo
- Preserve sourceAuthSecretId across skill re-imports/updates
- UI: Add PAT input field in import form for GitHub URLs
- UI: Add SkillAuthSection with ShieldCheck icon for viewing/updating/removing PAT
- UI: Add trash icon next to source label for delete-by-source
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
## Thinking Path
> - Paperclip orchestrates AI agents for zero-human companies
> - The issue detail page displays comment threads with rich timeline
rendering
> - Long threads (100+ items) cause severe typing lag in the comment
composer because every keystroke re-renders the entire timeline
> - CDP tracing confirmed 110ms avg key→paint latency and 60 long tasks
blocking the main thread for 3.7s total
> - This pull request memoizes the timeline, stabilizes callback props,
debounces editor observers, and reduces idle polling frequency
> - The benefit is responsive typing (21ms avg, 5.3× faster) even on
threads with 100+ timeline items
## What Changed
- **CommentThread.tsx**: Memoize `TimelineList` with `useMemo` so typing
state changes don't re-render 143 timeline items; extract
`handleFeedbackVote` to `useCallback`; added missing deps
(`pendingApprovalAction`, `onApproveApproval`, `onRejectApproval`) to
useMemo array
- **IssueDetail.tsx**: Extract inline callbacks (`handleCommentAdd`,
`handleCommentVote`, `handleCommentImageUpload`,
`handleCommentAttachImage`, `handleInterruptQueued`) to `useCallback`
with `.mutateAsync` deps (not full mutation objects) for stable
references; add conditional polling intervals (3s active / 30s idle) for
`liveRuns`, `activeRun`, `linkedRuns`, and timeline queries
- **MarkdownEditor.tsx**: Debounce `MutationObserver` and
`selectionchange` handlers via `requestAnimationFrame` coalescing
- **LiveRunWidget.tsx**: Accept optional `liveRunsData` and
`activeRunData` props to reuse parent-fetched data instead of duplicate
polling
## Verification
- Navigated to [IP address]:3105/PAPA/issues/PAPA-32 (thread with 100+
items)
- Typed in comment composer — lag eliminated, characters appear
instantly
- CDP trace test script (`test-typing-lag.mjs`) confirmed: avg 21ms
key→paint (was 110ms), 5 long tasks (was 60), 0.5s blocking (was 3.7s)
- Ran `pnpm test:run` locally — all tests pass
## Risks
- Low risk. All changes are additive memoization and callback
stabilization — no behavioral changes. Polling intervals are only
reduced for idle state; active runs still poll at 3–5s.
## Model Used
- Claude Opus 4.6 (`claude-opus-4-6`) via Claude Code CLI, with tool use
and extended context
## 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 run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] If this change affects the UI, I have included before/after
screenshots
- [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>
On resumed sessions, skipping --append-system-prompt-file (the original
fix) left two secondary issues:
- commandNotes still claimed the flag was injected, producing misleading
onMeta logs on every resumed heartbeat
- The instructions file was still read from disk and a combined temp file
written on every resume, even though effectiveInstructionsFilePath was
never consumed
Hoist canResumeSession before the I/O block and gate both the disk
operations and commandNotes construction on !canResumeSession / !sessionId.
Adds three regression tests: commandNotes is populated on fresh sessions,
empty on resume; and no agent-instructions.md is written on resume.
On resumed sessions the agent instructions are already present in the
session cache. Unconditionally passing --append-system-prompt-file
re-injects 5-10K redundant tokens per heartbeat and may be rejected by
the Claude CLI when combined with --resume.
Guard the flag behind `!resumeSessionId` so it is only appended on
fresh session starts.
Fixes: #2848
Pawla Abdul
Chris Farhood
dotta