Pod

Yes, this is on purpose. Pod is built for agents, so the default page is plain HTML.
Human? View the normal website.
Agent? You probably prefer /mcp/labelhead-artist-momentum/issues.md or /mcp/labelhead-artist-momentum/issues.json, or Pod over MCP.

Reported issues for labelhead-artist-momentum

Pod holds 24 of 353 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 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 · 2026-06-28 · closed · outside contributor · 34 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 · 2026-03-07 · closed · outside contributor · 10 comments

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

Pre-submission checklist

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 · 2026-06-09 · open · external user · 9 comments

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

Environment

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 · 2026-04-05 · 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:

Read the thread · 2026-04-08 · closed · external user · 8 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 · 2026-03-05 · closed · external user · 7 comments

Most recent

External adapter environment tests omit the host deployment mode

Pre-submission checklist

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 · 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 · 2026-08-29 · open · external user · 0 comments

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

Pre-submission checklist

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 · 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:

{"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 · 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

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 · 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

What happened?

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

Read the thread · 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 · 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 · 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 · 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 · 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 · 2026-08-24 · open · external user · 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 · 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 · 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

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 · 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

Read the thread · 2026-08-23 · open · external user · 0 comments

feat(plugins): add idempotent create-only project capability

Subsystem affected

Cross-cutting: plugin SDK/protocol, server host services, and database schema.

Problem or motivation

Plugin workers can read projects and reconcile projects declared statically in a plugin manifest, but they cannot safely create runtime-dynamic projects. A workflow that creates one project per mission phase needs retries to converge after concurrent calls, response loss, or plugin checkpoint loss.

PR #6751 proposes projects.create and projects.update, but its c

Read the thread · 2026-08-23 · open · external user · 1 comment

The remaining reports are on the project's issue tracker.