## Thinking Path
> - Paperclip orchestrates AI agents for zero-human companies
> - E2B is a sandbox provider plugin that runs agent code in isolated
cloud environments
> - Operators configure E2B through the plugin settings page
> - But the E2B API key configuration was unclear — the settings field
description didn't explain that pasted keys are auto-saved as company
secrets, and the fallback to the host `E2B_API_KEY` variable wasn't
documented
> - Additionally, long-running E2B sandbox commands were timing out
because the plugin environment RPC driver used a fixed timeout, and
environment commands competed for the single foreground command slot
> - This PR clarifies the E2B configuration UX, fixes RPC timeouts for
plugin environment execution, and runs E2B environment commands in
background mode to avoid blocking the foreground slot
> - The benefit is clearer E2B setup for operators and more reliable
sandbox command execution
## What Changed
- Updated E2B plugin manifest and settings UI to clarify API key
configuration — field description now explains that pasted keys are
saved as company secrets and documents the `E2B_API_KEY` host fallback
- Added test coverage for the plugin settings page rendering
- Fixed `plugin-environment-driver.ts` to pass the configured timeout
through to RPC calls instead of using a hardcoded default
- Updated `environment-runtime.ts` to propagate timeout from the
environment lease to the plugin driver
- Changed E2B sandbox command execution to use background handles so
long-running agent commands don't block the foreground slot needed by
the callback bridge
## Verification
- `pnpm test` — all existing and new tests pass
- `pnpm typecheck` — clean
- Manual: navigate to plugin settings, verify E2B API key field shows
the updated description text
- Manual: run an E2B-backed agent task with a long-running command,
verify it completes without RPC timeout
## Risks
- Low risk. Configuration UX change is cosmetic. The timeout fix passes
an existing value through instead of dropping it. Background command
execution is a behavioral change but only affects E2B sandbox commands —
the foreground slot is still available for bridge health checks.
## Model Used
Codex GPT 5.4 high via Paperclip.
## 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] 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
## Thinking Path
> - Paperclip orchestrates AI agents for zero-human companies
> - Agents run inside environments (local, SSH, E2B sandbox)
> - Operators need to configure and manage these environments
> - But environment settings were buried inside the general company
settings page, making them hard to find
> - Additionally, when testing an agent from the configuration form, the
test always ran locally regardless of which environment was selected
> - This PR moves environments into a dedicated top-level company
settings section and wires the "Test Environment" button to run inside
the selected environment
> - The benefit is operators can find and manage environments more
easily, and the test button now validates the actual environment the
agent will use
## What Changed
- Added a dedicated `CompanyEnvironments` settings page with its own
route and sidebar entry
- Updated `CompanySettingsSidebar` and `CompanySettingsNav` to include
the new environments section
- Modified the agent test route (`POST /agents/:id/test`) to accept an
optional `environmentId` parameter
- Updated all adapter `test.ts` handlers to resolve and use the
specified execution target environment
- Added `resolveTestExecutionTarget` to `execution-target.ts` for remote
environment test resolution with cwd fallback
- Moved the "Test Environment" button and its feedback display into the
`NewAgent` page footer for better UX flow
## Verification
- `pnpm test` — all existing and new tests pass
- `pnpm typecheck` — clean
- Manual: navigate to Company Settings, confirm "Environments" appears
as a top-level section
- Manual: configure an agent with a non-local environment, click "Test
Environment", confirm the test runs inside that environment
## Risks
- Low risk. UI-only routing change for the settings page. The
test-in-environment change adds an optional parameter with a local
fallback, so existing behavior is preserved when no environment is
specified.
## Model Used
Codex GPT 5.4 high via Paperclip.
## 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] 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
Per visual feedback the menu was sitting flush against the cursor;
introduce MENTION_MENU_CARET_GAP (10px) so the popover floats roughly
one space-width to the right of the caret.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
The board accepted the canonical executionDisposition payload + classifier
but found the visible badges in IssueRow / IssueProperties / Storybook too
noisy on production surfaces. UI-only walkback: keep the data layer, drop
every visible badge.
- Delete IssueDispositionBadge component and its test.
- Drop the badge import, suppression rule, and trailing-slot references in
IssueRow (desktop + mobile). Existing waiting pill, status-icon ring, and
blockerAttention copy remain unchanged.
- Drop the badge IIFE and related imports from IssueProperties so the
Status row only renders the StatusIcon again.
- Prune the badge cases from IssueRow.test (Live, Stalled, dispatchable,
Resuming, explicit-waiting suppression) so the file still type-checks.
- Storybook: remove "Canonical disposition badges" and "IssueRow with
disposition badge inline" sections plus DispositionRowSurface and
dispositionRowFixtures. The rest of Status Language is unchanged.
Kept: executionDisposition on issue payloads, the resolver and its test,
and the PAP-2818 contract doc — future features can consume the field
without re-running the classifier work.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Align CSS max-height with MENTION_MENU_HEIGHT (208) so the rendered
menu fills the layout-reserved space; previously max-h-[200px] clipped
the box 8px short of the JS layout budget, leaving a visible gap above
the caret line on the flipped path.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
- Drop the dispatchable "Ready" variant from the badge: the only place it ever
rendered was Storybook (IssueRow and IssueProperties already returned null
for it). dispositionCategory now returns null for dispatchable.
- Pull agent_continuable into a discrete "Resuming" category with a muted
indigo treatment so it doesn't share the rose alarm family with Recovery /
Needs attention / Stalled, and surface "Resuming · n/m" on the visible
badge instead of hiding the attempt counter in the tooltip.
- Differentiate human_escalation_required by owner on the visible label:
Needs board / Needs manager / Needs owner / Needs external. Color stays in
the rose family but the actionable bit is no longer hover-only.
- Bump invalid (Stalled) to a solid rose-600 fill so it stands out against
the rest of the rose family at scan distance, satisfying the parent plan's
acceptance criterion that the board can spot a stalled review leaf without
reading the full thread.
- Mirror the IssueRow suppression rule on IssueProperties so the Status row
doesn't double-render the same waiting/recovery signal as the chat-thread
IssueBlockedNotice.
- Add aria-label={tooltip} on the badge so screen readers get the path
detail, not just the category label.
- Storybook: rename and split the disposition matrix to cover all four
escalation owners, both Resuming attempts, and add a populated IssueRow
surface that shows the badge in production layout context (Live, Resuming,
Needs board, Needs manager, Stalled).
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Mention menu now sits on the same line as the caret and to the right of
the cursor (tracking each typed character instead of staying glued to
the @ marker). When the menu would overflow below the viewport it flips
above the line so it doesn't cover the line.
Touch handling no longer preventDefaults on touchstart, so the dropdown
scrolls natively on mobile; selection runs only on touchend if the
finger lifted with negligible movement (a tap, not a scroll).
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Adds IssueDispositionBadge that maps the canonical disposition to a
small palette of categories (live, ready, waiting, blocked-chain,
recovery, needs attention, stalled). Renders the badge in the inbox
row and on the issue detail status row, suppressing it where the
existing waiting/recovery copy already conveys the same signal so we
don't double up. Storybook gains a section showing every category for
UX review.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
The explicit-wait IssueBlockedNotice rendered Target/Owner/Resume as a row
of pills. Drop the Target pill entirely and render the remaining Owner and
Resume entries as a plain dl/dt/dd block — the value matters more than the
chip styling for these long-form details.
The "Jump to questions/confirmation/approval" affordance referenced anchor
ids that did not exist in the DOM (`issue-thread-interaction-{id}` and
`issue-approval-{id}`); the actual anchors are `interaction-{id}` and
`approval-{id}`. Switch to those ids and use a button with
scrollIntoView({behavior:"smooth"}) so the jump actually scrolls to the
target.
Add a small ExplicitWaitingJumpAnchor component that flips the arrow ↑/↓
based on whether the target is above or below the banner, recomputing on
scroll/resize.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
The Target/Owner/Resume chips in the explicit-wait IssueBlockedNotice
banner used inline-flex with `truncate max-w-[Nrem]` on inner spans, so
on narrow viewports a single chip (especially Resume's full sentence)
extended past the container edge. Cap each chip at max-w-full and let
the value span wrap on word boundaries; desktop layout is unchanged.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Wraps fenced code blocks in MarkdownBody with a relative container and an
absolute-positioned Copy button (top-right) that copies the block contents
to the clipboard via navigator.clipboard with an execCommand fallback for
non-secure contexts. The button is hidden by default and reveals on hover
or focus, swaps to a check icon and "Copied\!" label for ~1.5s after
clicking, and falls back to "Copy failed" when the clipboard write throws.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Replaces the rose liveness-break banner for issues whose blocker-attention
read model reports `state="covered"` with `reason="explicit_waiting"`.
- IssueBlockedNotice: add a sky-toned banner with Hourglass icon, kind-aware
headline/body copy (plan confirmation, user response, suggested tasks,
board approval, generic fallback), Target/Owner/Resume pill row, a "Jump
to confirmation" anchor, and a sky chip strip for parent-of-leaf chain
waits. Rose banner still wins on tie via the existing precedence.
- StatusIcon: add a sky ring + bottom-right nub for explicit_waiting and
emit `data-blocker-attention-state="explicit_waiting"`. Tooltip is
"Waiting on board/user/agent" or prefixed "Blocked · ..." when blocked.
- IssueRow: add a sky `Waiting · {owner}` pill (collapses to `Waiting` on
mobile) when the row's attention reason is explicit_waiting. Owner label
resolves via new optional agent/user maps wired in by IssuesList & Inbox.
- IssueDetail/IssueChatThread: lift the linkedApprovals query into the chat
tab and pass interactions/approvals/agentMap/userLabelMap/currentUserId
through to the new banner.
- Storybook + tests: extend coveredBlockedMatrix with sky cells, add four
new BlockedNotice fixtures (plan, user response, board approval, chain),
and cover the new states with unit tests on StatusIcon, IssueBlockedNotice,
and IssueRow.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip orchestrates AI agents through a company-scoped control
plane.
> - The affected surface is the board UI for issue threads, issue lists,
routines, dialogs, navigation, and issue review indicators.
> - Closed PR #4692 bundled backend, schema, docs, workflow, and UI/QoL
work into one oversized change set.
> - Greptile could not keep reviewing that broad PR because it exceeded
the 100-file review limit and mixed unrelated concerns.
> - This pull request extracts the UI/QoL slice into a fresh branch
under the review limit while leaving workflow and lockfile churn out.
> - The benefit is a focused review path for the board UI performance
and workflow improvements without reopening the oversized PR.
## What Changed
- Added long issue-thread virtualization, scroll-container binding,
anchor preservation, latest-comment jump targeting, and related
regression/perf fixtures.
- Improved issue list scalability with scroll-based loading, server
offset parameters, and pagination-focused UI tests.
- Reduced new issue dialog typing churn and split dialog action
subscriptions so broad layout/nav surfaces avoid unnecessary renders.
- Added routine variables help and routine description mention options
for users, agents, and projects.
- Added productivity review badge/link UI and fixed the badge to use
Paperclip's company-prefixed router link.
- Kept the split PR below Greptile's review limit and excluded
`.github/workflows/pr.yml` and `pnpm-lock.yaml`.
## Verification
- `pnpm install --no-frozen-lockfile` in the clean worktree to install
`@tanstack/react-virtual` locally without committing lockfile churn.
- `pnpm --filter @paperclipai/ui exec vitest run --config
vitest.config.ts src/components/IssueChatThread.test.tsx
src/components/IssuesList.test.tsx
src/components/NewIssueDialog.test.tsx src/pages/Routines.test.tsx
src/pages/Issues.test.tsx` passed: 5 files, 83 tests.
- `pnpm --filter @paperclipai/ui typecheck` passed.
- `git diff --check origin/master..HEAD` passed.
- Split-scope checks: 53 changed files; no `.github/workflows/pr.yml`;
no `pnpm-lock.yaml`.
- Screenshots were not captured in this heartbeat; the changes are
primarily virtualization, routing, pagination, and editor behavior
covered by focused regression tests.
## Risks
- Moderate UI risk because issue-thread virtualization changes scroll
behavior on long conversations; regression tests cover anchor jumps,
latest-comment targeting, row metadata, and short-thread fallback.
- Moderate integration risk because the issue-list offset parameter and
productivity review field depend on matching API behavior.
- Dependency risk: the UI package adds `@tanstack/react-virtual` while
repository policy keeps `pnpm-lock.yaml` out of PRs, so CI must resolve
dependency changes through the repo's normal lockfile policy.
> 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 local repository and
GitHub workflow. Exact runtime context window was not exposed by the
harness.
## 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
- [ ] 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>
The user reported having to click Jump-to-latest ~10 times before the
viewport actually pinned to the latest comment on long threads (e.g.
PAP-2536). The virtualizer estimates 220px per unmeasured row, so
totalSize is hugely underestimated until rows render and get measured;
a single smooth scroll lands above the real bottom and each subsequent
click walks closer as more rows mount and expand.
- IssueChatThread.handleJumpToLatest now awaits an optional
onRefreshLatestComments before resolving the latest target so any
comment that arrived after the initial load (or was dropped on a
live-update reconnect) is in the loaded set first.
- scrollToLatestCommentWithSettle issues the initial smooth scroll,
then on an 80 ms tick re-issues scrollIntoView({ block: "end" }) on
the latest comment row for up to 4 s, exiting once the row's bottom
matches the scroll container's bottom (within 4 px) and scrollTop +
scrollHeight have been stable for 3 ticks. If the row hasn't been
mounted into the virtualizer buffer, scrollToIndex is used to nudge
the offset until it is.
- IssueDetail destructures `refetch` from the comments useInfiniteQuery
and threads refetchLatestComments through to IssueChatThread.
QA verified on PAP-2720: single-click convergence in 327-407 ms across
5 successive trials, comment-fa353646 lands flush with the viewport
bottom (-0.44 px sub-pixel), stopped-scroll stability holds.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
## Thinking Path
> - Paperclip is the control plane for autonomous AI companies, so
backend task ownership, recovery, review visibility, and company-scoped
limits need to stay enforceable without UI-only coupling.
> - Closed PR #4692 bundled those backend changes with UI workflow,
docs, skills, workflow, and lockfile churn.
> - PAP-2694 asks for a clean backend/control-plane slice from that
closed branch.
> - This branch starts from current `master` and mines only the `cli`,
`packages/db`, `packages/shared`, and `server` contracts/tests needed
for the backend behavior.
> - It explicitly excludes UI workflow/performance work,
`.github/workflows/pr.yml`, `pnpm-lock.yaml`, docs, skills,
package-script, adapter UI build-config, and perf fixture script
changes; the only UI files are fixture/test updates required by the
tightened shared `Company` contract.
> - The benefit is a smaller reviewable PR that preserves the
control-plane fixes while staying under Greptile s 100-file review
limit.
## What Changed
- Added company-scoped attachment-size limits through DB
schema/migrations, shared company portability contracts, CLI
import/export coverage, and server attachment upload enforcement.
- Added productivity review service/API behavior for no-comment streak,
long-active, and high-churn review issues, including request-depth
clamping and issue summary exposure.
- Hardened issue ownership and recovery/control-plane paths: peer-agent
mutation denial, issue tree pause/resume behavior, stranded recovery
origins, and related activity/test coverage.
- Preserved related backend contract updates for routine timestamp
variables and managed agent instruction bundles because they live in
shared/server contracts from the source branch.
- Addressed Greptile feedback by making `Company.attachmentMaxBytes`
non-optional, simplifying review request-depth clamping, fixing the
migration final newline, and enforcing the process-level attachment cap
as the final ceiling for uploads.
- Added minimal company fixtures needed for repo-wide typecheck/build
and kept the PR to 66 changed files with forbidden/non-slice paths
excluded.
## Verification
- `pnpm install --frozen-lockfile`
- `git diff --check origin/master..HEAD`
- `git diff --name-only origin/master..HEAD | wc -l` -> 66 files
- `git diff --name-only origin/master..HEAD -- .github/workflows/pr.yml
pnpm-lock.yaml package.json doc skills .agents scripts
packages/adapters` -> no output
- `pnpm exec vitest run --config vitest.config.ts
packages/shared/src/validators/issue.test.ts
packages/shared/src/routine-variables.test.ts
packages/shared/src/adapter-types.test.ts
cli/src/__tests__/company-import-export-e2e.test.ts
cli/src/__tests__/company.test.ts
server/src/__tests__/productivity-review-service.test.ts
server/src/__tests__/issue-tree-control-service.test.ts
server/src/__tests__/issue-tree-control-routes.test.ts
server/src/__tests__/issue-agent-mutation-ownership-routes.test.ts
server/src/__tests__/issue-attachment-routes.test.ts
server/src/__tests__/heartbeat-process-recovery.test.ts
server/src/__tests__/issues-service.test.ts` -> 12 files, 147 tests
passed
- `pnpm exec vitest run --config vitest.config.ts
cli/src/__tests__/company-delete.test.ts
cli/src/__tests__/company-import-export-e2e.test.ts
server/src/__tests__/productivity-review-service.test.ts` -> 3 files, 18
tests passed
- `pnpm exec vitest run --config vitest.config.ts
server/src/__tests__/issue-attachment-routes.test.ts` -> 1 file, 6 tests
passed
- `pnpm --filter @paperclipai/db typecheck && pnpm --filter
@paperclipai/shared typecheck && pnpm --filter @paperclipai/server
typecheck && pnpm --filter paperclipai typecheck`
- `pnpm --filter @paperclipai/server typecheck`
- `pnpm --filter @paperclipai/ui typecheck && pnpm --filter
@paperclipai/ui build`
## Risks
- Includes migrations `0073_shiny_salo.sql` and
`0074_striped_genesis.sql`; merge ordering matters if another PR adds
migrations first.
- This is intentionally backend-only apart from fixture/test updates
forced by shared type correctness; UI affordances from PR #4692 are not
present here and should land in separate UI slices.
- The worktree install emitted plugin SDK bin-link warnings for unbuilt
plugin packages, but the targeted tests and package typechecks completed
successfully.
> 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 terminal/GitHub
workflow. Exact runtime context window was not exposed by the harness.
## 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] 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>
StatusIcon adds an AlertTriangle overlay on the red-ring blocked variant
when the read model reports recovery_needed; tooltip and aria-label
spell out the liveness break and which leaf is invalid.
IssueBlockedNotice gets a rose container variant with two copy
templates: parent-perspective ("paused at a liveness break in [leaf]")
and leaf-perspective ("productive run that exited without queueing a
continuation"). The sub-row label flips to "Liveness break at" and an
owner pill renders with the next-action verb derived from the
nextActionHint.
IssueRow surfaces "liveness break at {leaf}" in the meta line for
recovery_needed parents so inbox/MyIssues rows show the actionable leaf
without opening the parent.
The status-language Storybook matrix gains the four new state/reason
combinations and parent/leaf-perspective notice variants. Targeted
vitest cases cover both the icon overlay and notice rendering.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
The Phase 3 virtualizer (PAP-2645) used useWindowVirtualizer, but on the
real issue page the chat lives inside `<main id="main-content"
overflow-auto>`. Window scroll never moves on desktop, so the virtualizer
only ever mounted the first 7 rows and the chat area went blank past the
first viewport. Detect the actual scroll container at mount via DOM walk
and rekey the inner virtualizer to use element-mode observers/scroll-fn
when an overflow ancestor exists; fall back to window mode for mobile and
the auth-free perf fixture, where the document scrolls.
Adds a regression test that mounts IssueChatThread inside an overflow-auto
ancestor and asserts jump-to-latest drives that element's scrollTo, not
window.scrollTo.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
If the issue reply composer would post without any assignee, the first
Send press now surfaces a warning toast instead of submitting. A second
press confirms the user wants to proceed unassigned, and changing the
assignee resets the confirmation.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Add @tanstack/react-virtual and gate IssueChatThread on a 150-row
threshold. Thread shorter than the threshold still uses the direct
render path; longer threads window the merged feed via
useWindowVirtualizer with dynamic measurement and overscan, so the
document scroll continues to represent the full conversation height
without introducing a nested scrollbar.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
UX feedback split the badge by role: on the source issue the linked pill
describes a state ("Under review"), while on the productivity-review issue
itself the static pill identifies the kind of issue ("Productivity review").
Tooltip and aria-label still surface the formal name on hover.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
Adds a productivityReview lookup to single-issue and list responses, plus
a ProductivityReviewBadge component that surfaces the open productivity
review on the source issue's header (with trigger/no-comment-streak
detail in the tooltip), and a compact eye glyph next to the StatusIcon
on IssueRow so operators can spot yellow tasks in any list. The review
issue itself shows a static "Productivity review" pill in the header
when its originKind matches.
Also extends the shared IssueOriginKind union with the recovery-side
kinds (harness_liveness_escalation, issue_productivity_review,
stranded_issue_recovery) so UI checks against issue.originKind type-check.
Wires the new helper through GET /issues/:id, GET /companies/:id/issues
(via issueService.list), and GET /issues/:id/heartbeat-context. Adds a
status-language Storybook story matrix for UX review.
Co-Authored-By: Paperclip <noreply@paperclip.ing>
