# Reported issues for io.github.IncorporatedPartners/labelhead-artist-momentum

Pod holds 24 of 354 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded entirely — a maintainer's release checklist is not a warning to a prospective user.

Back to [io.github.IncorporatedPartners/labelhead-artist-momentum](/mcp/io-github-incorporatedpartners-labelhead-artist-momentum).

## Most discussed

### bug: hermes-local and process adapters don't forward onSpawn, causing false-positive orphan reaps

## Description

The `hermes-local` and `process` adapters don't forward `ctx.onSpawn` to `runChildProcess()`, so `processPid` is never persisted to the `heartbeat_runs` DB table.

All other adapters (claude-local, codex-local, cursor-local, gemini-local, grok-local, opencode-local, pi-local) already forward `ctx.onSpawn` — these two are the only ones missing it.

## Impact

Without PID persistence, the orphan reaper cannot distinguish live runs from abandoned processes. When a run's `updatedAt` 

[Read the thread](https://github.com/paperclipai/paperclip/issues/8723) · 2026-06-28 · closed · outside contributor · 34 comments

### execution_workspaces never auto-close — stale active lifecycle leak (no terminal hook / cascade / reaper)

### Pre-submission checklist

- [x] I have searched existing open and closed issues and this is not a duplicate.
- [x] I am on the latest released version of Paperclip (or can reproduce on `master`).
- [x] I have confirmed the error originates in Paperclip itself — not in my agent adapter, API provider, or local configuration.

### What happened?

`execution_workspaces` (ews) rows are never automatically closed when the work that owns them ends. The columns `closed_at` / `cleanup_eligible_at` ar

[Read the thread](https://github.com/paperclipai/paperclip/issues/7790) · 2026-06-09 · open · external user · 9 comments

### Skills Catalog tab returns 500: server resolves manifest via monorepo path not present in published build

## Summary

On a fresh global install of `paperclipai` (latest, `2026.529.0`), the **Skills → Catalog** tab fails with `Internal server error`. The API endpoint `GET /api/skills/catalog` returns **500** because the server looks for the skills-catalog manifest at a monorepo-relative path that does not exist in the published npm layout.

## Error

```
ERROR: GET /api/skills/catalog 500 — Skills catalog manifest not found at
/Users/<user>/.npm/_npx/<hash>/node_modules/@paperclipai/packages/skills-c

[Read the thread](https://github.com/paperclipai/paperclip/issues/7281) · 2026-05-31 · closed · external user · 8 comments

### Agent completion comments can be attributed to the user and retrigger the assignee (causing status drift, duplicate comments, and agent run loops)

### 🐞 **Bug Description**

When an agent-assigned task completes in Paperclip, the agent generates the appropriate completion markdown comment (e.g., review notes or artifact links). However, this comment sometimes appears as if it was authored by the logged-in human user (board operator), not the agent. This misattribution leads to a buggy workflow where:

- The comment is shown in the UI as from “You” rather than the expected agent name.
- Posting of this (misattributed) comment immediately re

[Read the thread](https://github.com/paperclipai/paperclip/issues/3817) · 2026-04-16 · closed · external user · 8 comments

### openclaw_gateway run fails: invalid agent params due to unexpected root-level paperclip field

## Summary

Paperclip `openclaw_gateway` can now connect successfully to OpenClaw after fixing LAN reachability and device pairing scopes, but actual runs fail because the adapter sends a root-level `paperclip` field that the current OpenClaw agent endpoint rejects.

Current error:

```text
[openclaw-gateway] request failed: invalid agent params: at root: unexpected property 'paperclip'
```

## What was fixed before this error

We previously resolved these issues:

- Paperclip container could re

[Read the thread](https://github.com/paperclipai/paperclip/issues/3089) · 2026-04-08 · closed · external user · 8 comments

### Gemini adapter probe fails with "Failed to relaunch the CLI process" on Gemini CLI v0.36.0

## Environment
- Gemini CLI version: 0.36.0
- OS: Windows 11 + WSL2 (Ubuntu)
- Paperclip version: latest master

## Problem
The adapter environment check fails with:
> Fatal error: Failed to relaunch the CLI process

Even when `gemini` is installed and working correctly in the terminal.

## Root Causes Found (after debugging)

1. **Trust dialog blocks subprocess** — Gemini CLI v0.36.0 added a folder trust 
   prompt on first run in any directory. When Paperclip spawns gemini as a 
   non-interac

[Read the thread](https://github.com/paperclipai/paperclip/issues/2852) · 2026-04-05 · open · external user · 9 comments

### PROPOSAL: Sandboxed Agent Execution — provider-agnostic interface for secure remote environments

## Summary

Paperclip needs a way to run agents in sandboxed, isolated environments to safely process untrusted input (e.g. GitHub issues, PRs) without risk of prompt injection compromising the host or leaking secrets.

This proposal introduces a **provider-agnostic sandbox interface** that lets operators bring their own sandbox runtime (E2B, Cloudflare Containers, Daytona, Fly.io Sprites, etc.) while Paperclip handles orchestration, auth, and lifecycle management.

## Problem

Running agents th

[Read the thread](https://github.com/paperclipai/paperclip/issues/248) · 2026-03-07 · closed · outside contributor · 10 comments

### MCPs?

It would be nice to have Agents able to call MCPs or tools like Figma. Is this something on the roadmap?

[Read the thread](https://github.com/paperclipai/paperclip/issues/50) · 2026-03-05 · closed · external user · 7 comments

## Most recent

### External adapter environment tests omit the host deployment mode

## Pre-submission checklist

- [x] I searched open and closed issues and pull requests. I did not find this report.
- [x] I reproduced this problem on the current `master` branch.
- [x] I confirmed that the supported Paperclip route omits data that is already in the public adapter context type.

**What happened?**

`POST /api/companies/:companyId/adapters/:adapterType/test-environment` calls an external adapter with an `AdapterEnvironmentTestContext`. The route sends the company, adapter type, c

[Read the thread](https://github.com/paperclipai/paperclip/issues/12481) · 2026-08-29 · open · external user · 0 comments

### Bootstrap @paperclipai/adapter-aider-local on npm (one-time placeholder publish)

## What needs to happen

A maintainer with `@paperclipai` npm scope access runs the repo's own bootstrap helper once:

```
node scripts/bootstrap-npm-package.mjs @paperclipai/adapter-aider-local --publish
```

Per the helper's header this publishes an empty 0.0.0 placeholder that only reserves the name — real package content is only ever published by CI after review and merge, the placeholder needs no local build or workspace state, and it can be run from any checkout including master. The first

[Read the thread](https://github.com/paperclipai/paperclip/issues/12480) · 2026-08-29 · open · external user · 0 comments

### v2026.824.1: Kubernetes Sandbox catalog install returns 200 but persists no plugin

### Pre-submission checklist

- [x] I have searched existing open and closed issues and this is not a duplicate.
- [x] I am on the latest released version of Paperclip (or can reproduce on `master`).
- [x] I have confirmed the error originates in Paperclip itself — not in my agent adapter, API provider, or local configuration.

### What happened?

On the official `ghcr.io/paperclipai/paperclip:2026.824.1` image, installing the first-party Kubernetes Sandbox (alpha) catalog plugin returns HTTP 20

[Read the thread](https://github.com/paperclipai/paperclip/issues/12430) · 2026-08-28 · open · external user · 0 comments

### heartbeat-runs cancel: OpenAPI advertises agent auth that doesn't exist (same bug class as #9726/#11946)

## What happened?

`POST /api/heartbeat-runs/{runId}/cancel` (`server/src/routes/agents.ts`) opens with
`assertBoard(req)` — unconditional board-only, no agent bearer path at all. A valid
agent bearer token plus the current `X-Paperclip-Run-Id` gets exactly:

```json
{"error":"Board access required"}
```

But `/api/openapi.json` advertises `x-paperclip-authorization.actor: "board_or_agent"`
and includes `AgentBearerAuth` security for this operation. Confirmed by reading the
generator: `resolveOp

[Read the thread](https://github.com/paperclipai/paperclip/issues/12411) · 2026-08-28 · open · external user · 0 comments

### local-process sandbox proxy: network_target_denied reason is unreachable on the CONNECT path

### Pre-submission checklist

- [x] I have confirmed this improves existing behavior — it does not add a new command, endpoint, or concept.
- [x] I have searched existing open and closed issues and this has not already been proposed.

### What existing behavior does this improve?

The local-process sandbox egress proxy's `403` denial response on the **CONNECT** path
(`connectProxyError()` in `packages/adapter-utils/src/local-process-sandbox.ts`).

### Subsystem affected

packages/adapters — agen

[Read the thread](https://github.com/paperclipai/paperclip/issues/12328) · 2026-08-27 · open · external user · 0 comments

### local-process sandbox: --dev /dev omitted for network-only scopes, breaking /dev/null inside the sandbox

### Pre-submission checklist

- [x] I have searched existing open and closed issues and this is not a duplicate.
- [x] I am on the latest released version of Paperclip (or can reproduce on `master`).
- [x] I have confirmed the error originates in Paperclip itself — not in my agent adapter, API provider, or local configuration.

### What happened?

`buildLocalProcessSandboxSpawnTarget` emits `--dev /dev` only inside the
`filesystemScope === "workspace"` branch. When a sandbox is constructed with 

[Read the thread](https://github.com/paperclipai/paperclip/issues/12327) · 2026-08-27 · open · external user · 0 comments

### MCP gateway rejects agent tokens ("did not verify") — tool_gateway_sessions never populated; all governed tools absent to agents

### Summary

On a self-hosted instance (embedded PostgreSQL, `claude-local` adapter), agents cannot use **any** governed MCP gateway tools. For every run the adapter writes a complete `mcp-config.json` with the connected MCP servers as `type: "http"` entries pointing at `/api/tool-gateway/gateways/<id>/mcp` with a `pcgw_…` bearer token — but every request to that endpoint returns:

```
HTTP 401 {"error":"Agent token did not verify; obtain fresh credentials and retry"}
```

so the MCP servers fai

[Read the thread](https://github.com/paperclipai/paperclip/issues/12294) · 2026-08-27 · closed · outside contributor · 2 comments

### gemini_local is unusable in a containerized/headless local deployment (three independent blockers)

# `gemini_local` is unusable in a containerized/headless local deployment (three independent blockers)

## Summary

Running the official Docker image on a single host, `gemini_local` fails at three
separate stages before a single turn can execute. Each has a distinct root cause,
and each is a one-to-few-line fix. All three share one assumption: that an
`executionTarget` of *local* implies an *interactive* environment. In a container
that is false — local is exactly as headless as remote.

The th

[Read the thread](https://github.com/paperclipai/paperclip/issues/12233) · 2026-08-26 · open · external user · 0 comments

### opencode_local: server PWD leaks into child env, OpenCode resolves wrong project root (silent wrong-checkout runs)

### What happened?

The `opencode_local` adapter spawns `opencode run` with the correct `cwd` but an env inherited from the **Paperclip server process**. Node's `spawn` sets the OS working directory from `cwd` but does **not** rewrite `PWD` in the supplied `env`, so the child keeps the server's `PWD` (e.g. `/Users/REDACTED/workspace/paperclip`, the checkout the server was started from).

OpenCode **>= 1.14.48 resolves its project/worktree root from the `PWD` env var, not from `process.cwd()`**. 

[Read the thread](https://github.com/paperclipai/paperclip/issues/12220) · 2026-08-26 · open · external user · 0 comments

### gemini_local ACP adapter never sends authenticate for oauth-personal (Google account login)

## Bug: Gemini ACP adapter never authenticates via `oauth-personal` (Google account login), even when valid OAuth credentials exist

**Version:** paperclipai 2026.824.0

**Summary**

The `gemini_local` adapter's ACP execution path never sends the ACP `authenticate` RPC for Gemini CLI's `oauth-personal` auth method ("Log in with Google"), even when the host already has a valid, non-expired OAuth session (`~/.gemini/oauth_creds.json` + `~/.gemini/settings.json` with `selectedType: "oauth-per

[Read the thread](https://github.com/paperclipai/paperclip/issues/12183) · 2026-08-25 · open · external user · 0 comments

### Feature request: vendor-neutral async gateway adapter (the `hermes_gateway` run/SSE contract, unbranded)

## The ask

Paperclip already has a good async agent-wake protocol. The `hermes_gateway` adapter creates a run with `POST /v1/runs`, observes it over `GET /v1/runs/{id}/events`, polls `GET /v1/runs/{id}` as a fallback, and calls `POST /v1/runs/{id}/stop` on timeout. That shape is genuinely runtime-agnostic — nothing about "create a run, stream its events, stop it" depends on which agent runtime is on the other end. But it is currently reachable only through an adapter named for one specific runt

[Read the thread](https://github.com/paperclipai/paperclip/issues/12110) · 2026-08-24 · open · external user · 0 comments

### Plugin jobs carry no company scope — under governed access, scheduled jobs cannot do scoped reads without plugin-side ownership tracking

## Summary

Since #9557 (governed access), `config.get` and `secrets.resolve` require a company scope (`resolveRequiredCompanyId` in `packages/plugins/sdk/src/host-client-factory.ts` throws `InvocationScopeDeniedError` otherwise). Every host→plugin surface carries that scope except one:

- events: `PluginEvent.companyId` — required (`packages/plugins/sdk/src/types.ts`)
- tools: `ToolRunContext.companyId` — required
- actions: `actorContext.companyId`
- **jobs: `PluginJobContext = { jobKey, runId

[Read the thread](https://github.com/paperclipai/paperclip/issues/12109) · 2026-08-24 · open · outside contributor · 0 comments

### DX: adapter_failed surfaces only the HTTP status, not the endpoint or the config that produced it

### Summary

When a `claude_local` agent run fails, the run is marked `adapter_failed` and the surfaced message is only the upstream HTTP status. It contains nothing about **which endpoint** was called or **which configuration** produced it, so the actual cause is not discoverable from Paperclip.

### What we saw

```
adapter_failed - Claude run failed: subtype=success: API Error: 405 Not Allowed
```

Every `claude_local` run failed this way, on every agent, for weeks.

### The actual cause

`~/

[Read the thread](https://github.com/paperclipai/paperclip/issues/12105) · 2026-08-24 · open · external user · 0 comments

### feature: agent-attributed JWT for automation actors (avoid comment-reopen loop on stranded issues)

## Problem

When Paperclip's recovery / automation layer closes a stranded heartbeat issue by posting a "closing" comment, the comment is stamped with the local-board JWT — which attributes as `user`. There's an implicit rule that user comments on a closed issue reopen it. Result: automation's own close-comment reopens the issue it just closed. Infinite loop.

Observed in our environment as a chronic pattern on stranded-HB recovery flows (issues in `blocked` status with `activeRecoveryAction.kin

[Read the thread](https://github.com/paperclipai/paperclip/issues/12086) · 2026-08-24 · open · external user · 0 comments

### A failed runtime MCP gateway handshake records nothing and reaches no agent — tool loss is indistinguishable from "not provisioned"

### Pre-submission checklist

- [x] I have confirmed this improves existing behavior — it does not add a new command, endpoint, or concept.
- [x] I have searched existing open and closed issues and this has not already been proposed.

### What existing behavior does this improve?

Runtime MCP gateway delivery: `buildPaperclipRuntimeMcpServers` and `recordRuntimeMcpDeliveryDiagnostic` in `server/src/services/heartbeat.ts`.

### Subsystem affected

server/ — REST API & orchestration services

### 

[Read the thread](https://github.com/paperclipai/paperclip/issues/12076) · 2026-08-24 · open · external user · 0 comments

### pi_local model discovery: stderr || stdout heuristic breaks when extensions emit stderr noise

## Summary

`pi_local` model discovery returns `"Pi returned no models"` when the spawned pi process writes anything to stderr besides the model table — e.g. diagnostics from a pi extension or wrapper script. The fix in #1787 made the parser prefer `result.stderr`, which breaks for any pi extension that emits stderr output.

## Environment

- paperclipai: **2026.722.0**
- pi: **v0.84.2**
- adapter: `@paperclipai/adapter-pi-local` (bundled with paperclipai)
- OS: Linux (Ubuntu 24.04)
- A pi exten

[Read the thread](https://github.com/paperclipai/paperclip/issues/12056) · 2026-08-23 · open · external user · 0 comments

The remaining reports are on [the project's issue tracker](https://github.com/paperclipai/paperclip/issues).