Menu
Extract non-obvious learnings from session to AGENTS.md files to build codebase understanding
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Add a new persistent self-hosted service on sietch, exposed on the tailnet as <name>.sole-pierce.ts.net via Tailscale Services. Use when the user asks to host, self-host, deploy, or expose a new service (web app, API, MCP server, media tool) on sietch or "my server". Covers the nix module, agenix secrets, the tailscale admin-console steps, and verification.
Check and update digest pins for third-party docker images used by selfhost services on sietch. Use when the user asks to update, bump, or check container images, image pins, or docker digests for self-hosted services.
Desktop inspection and control for the local NixOS + niri Wayland session. Use when an agent needs to see or drive the desktop - take screenshots of windows/outputs (including hidden workspaces), inspect windows/workspaces via niri, focus or launch apps (with or without stealing focus), click/type/scroll into GUI apps, run app tasks in the background on the agent workspace, or recover a confused desktop (overview open, focus lost). Route browser work to agent-browser/CDP and terminal work to tmux; use this skill's acu tool for everything the compositor and pixels must solve.
Control a dedicated Helium browser profile through agent-browser and Chrome DevTools Protocol. Use when an agent needs to drive the user's Helium browser locally or through an SSH-forwarded CDP endpoint, inspect browser state, open pages in a visible browser, collaborate on local or remote dev-server debugging, take screenshots, read snapshots, manage tabs, or validate browser behavior in the Helium profile.
Ask Codex CLI (gpt-5.5) to run local app verification that needs browser automation, desktop control, screenshots, app launching, or independent runtime inspection. This is how gpt-5.5 is invoked for computer-use work, and the preferred route for browser verification on this machine. Use when the user asks to test a flow, verify UI behavior, inspect a running app, capture screenshots, or report confirmation and feedback about implemented behavior.
Ask Codex CLI (gpt-5.5) to implement scoped code changes in the current repository, then have the current agent inspect the resulting diff and verification. This is how gpt-5.5 is invoked for implementation work. Use when the user asks to delegate implementation to Codex or gpt-5.5, when the model-selection rubric routes the work to gpt-5.5, or when a bounded task would benefit from another coding agent producing a patch.
| name | learn |
| description | Extract non-obvious learnings from session to AGENTS.md files to build codebase understanding |
| context | fork |
Analyze this session and extract non-obvious learnings to add to AGENTS.md files.
AGENTS.md files can exist at any directory level, not just the project root. When an agent reads a file, any AGENTS.md in parent directories are automatically loaded into the context of the tool read. Place learnings as close to the relevant code as possible:
What counts as a learning (non-obvious discoveries only):
What NOT to include:
Process:
ln -s AGENTS.md CLAUDE.mdAfter updating, summarize which AGENTS.md files were created/updated and how many learnings per file.
$ARGUMENTS