Files

T

Chris Farhood 2fd9f0691d Add dev namespace access and kubectl to POLICIES.md and TOOLS.md

Each org now has a -dev namespace where agents can freely use kubectl
for testing and iteration. Production namespaces remain Flux-only.

Access model:
- Cluster-wide: read-only
- Production namespace: read-write (Flux-managed, no manual kubectl)
- Dev namespace: read-write (agents may use kubectl freely)

Co-Authored-By: Paperclip <noreply@paperclip.ing>

2026-03-21 11:05:32 -04:00

2.0 KiB

Raw Blame History

Privileged Escalation — Shared Tools

GitHub Authentication

export GH_TOKEN=$(bash /paperclip/privilegedescalation/agents/get-github-token.sh)

Run this at the start of every heartbeat. Sets GH_TOKEN for gh and git.

Paperclip API

Auto-injected env vars:

PAPERCLIP_API_URL — base URL (fall back to http://localhost:3100)
PAPERCLIP_API_KEY — short-lived JWT for this run
PAPERCLIP_RUN_ID — include on all mutating requests

Available Tools

Tool	Purpose
`gh`	GitHub CLI — issues, PRs, CI runs, repo management
`git`	Version control — branches, commits, PRs
`curl`	HTTP requests — Paperclip API, external services
`jq`	JSON parsing and formatting
`node` / `npm` / `pnpm` / `npx`	Node.js runtime and package management
`python3`	Python scripting
`pnpm paperclipai`	Paperclip CLI — issue/agent operations
`kubectl`	Kubernetes CLI — read-only cluster-wide, read-write in `privilegedescalation` and `privilegedescalation-dev`

Repos

Repo	Owner	Purpose
`privilegedescalation/agents`	Board	Agent profiles and configuration (this repo)
`privilegedescalation/headlamp-*`	Gandalf	Headlamp plugin repos

MCP Servers

Server	Endpoint	Available To	Purpose
`minimax-search`	Local (uvx)	VP Product, CMO	Web search and image understanding
`playwright-privilegedescalation`	`http://playwright-privilegedescalation.paperclip.svc.cluster.local:3000/sse`	Regression Regina (QA)	Playwright browser automation for E2E testing

MCP server configs live in each agent's .mcp.json (claude_local) or opencode.json (opencode_local).

GitHub Actions Runners

Self-hosted ARC runners are available at the org level. Use runs-on: runners-privilegedescalation in workflows.

Runners scale to zero when idle — if no runner pods are visible, they will start automatically when a workflow is triggered.

2.0 KiB Raw Blame History