feat!: turn the skills repo into agent-env (m4rti.agents module) #2

Merged

marti merged 4 commits from feature/agent-env into main 2026-10-13 19:42:03 +02:00

Conversation 0 Commits 4 Files changed 30 +2299 -205

marti commented 2026-10-13 19:38:01 +02:00

Owner

• Copy link
• Copy Markdown

Summary

The skills repo becomes the agent-env repo: one Home Manager module, m4rti.agents, replaces m4rti-skills and the scattered agent config in the NixOS repo. apps decides which agents exist; every other part is installed into whichever enabled agents can take it.

m4rti.agents = {
  apps = {
    claude.code.enable = true;  claude.desktop.enable = true;
    codex.enable = true;        gpt.desktop.enable = true;
    opencode.cli.enable = true; opencode.desktop.enable = true;
    pi.enable = true;           t3code.enable = true;
  };
  skills.enable = true;     rules.enable = true;
  mcp.enable = true;        plugins.enable = true;
  subagents.enable = true;
  providers.cumsek = { url = "https://ai.cumsek.com"; keyFile = osConfig.sops.secrets."...".path; };
};

 nix/
-├── hm-module.nix        # m4rti-skills
+├── module/
+│   ├── apps.nix         # the agents and their default settings
+│   ├── skills.nix       # skills/ -> every agent
+│   ├── rules.nix        # rules/ -> rulesDir / AGENTS.md
+│   ├── mcp.nix          # one server list, secrets read from files at startup
+│   ├── plugins.nix      # ponytail, graft, open-code-review, zeit, agent-browser
+│   ├── subagents.nix    # subagents/ -> each agent's format
+│   └── providers.nix    # LiteLLM gateways for opencode and pi
+├── subagents.nix        # frontmatter parser + per-agent renderers
 └── ponytail.nix
+rules/                   # code-style, subagents (new)
+subagents/               # 10 agents: 5 rewritten, 5 new
+providers/               # gateway model discovery scripts

How each agent reads a secret (env or headers value { file = ...; }) at startup:

Agent	stdio env	HTTP headers
Claude Code	wrapper script	headersHelper
codex	wrapper script	env_http_headers + codex wrapper
opencode	{file:...}	{file:...}
pi	!cat	!cat

Evidence

• Before: nix flake check covered skill frontmatter and lint only; the module was never evaluated.
  After: new subagents check (parses and renders all 10) and module check (full Home Manager evaluation, every part on, secret-backed stdio + HTTP MCP servers) -> all checks passed!
• Built and inspected the generated pieces: the Claude headersHelper emits {"X-Api-Key": "<file contents>"}; the codex wrapper contains export MCP_HTTP_SECRET_X_API_KEY="$(cat /run/secrets/key)"; read-only subagents render with sandbox_mode = "read-only" (codex), disallowedTools (Claude), tools: {...: false} (opencode), a reduced allowlist (pi).
• Not done: a real switch on a host, or launching each agent against the generated config.

Merge Danger

Door: two-way

Nothing consumes this until the NixOS config is migrated; reverting the merge restores m4rti-skills.

Blast Radius: consumers

m4rti-skills.* is removed, so the NixOS config fails to evaluate on the next nix flake update skills until it moves to m4rti.agents. When migrating: drop the CODEX_HOME session variable (it would hide the Home Manager codex config), drop the separate zeit module import (this module imports it), and note Claude Code now defaults to the llm-agents build on every host.

🤖 Generated with Claude Code

## Summary The skills repo becomes the agent-env repo: one Home Manager module, `m4rti.agents`, replaces `m4rti-skills` and the scattered agent config in the NixOS repo. `apps` decides which agents exist; every other part is installed into whichever enabled agents can take it. ```nix m4rti.agents = { apps = { claude.code.enable = true; claude.desktop.enable = true; codex.enable = true; gpt.desktop.enable = true; opencode.cli.enable = true; opencode.desktop.enable = true; pi.enable = true; t3code.enable = true; }; skills.enable = true; rules.enable = true; mcp.enable = true; plugins.enable = true; subagents.enable = true; providers.cumsek = { url = "https://ai.cumsek.com"; keyFile = osConfig.sops.secrets."...".path; }; }; ``` ```diff nix/ -├── hm-module.nix # m4rti-skills +├── module/ +│ ├── apps.nix # the agents and their default settings +│ ├── skills.nix # skills/ -> every agent +│ ├── rules.nix # rules/ -> rulesDir / AGENTS.md +│ ├── mcp.nix # one server list, secrets read from files at startup +│ ├── plugins.nix # ponytail, graft, open-code-review, zeit, agent-browser +│ ├── subagents.nix # subagents/ -> each agent's format +│ └── providers.nix # LiteLLM gateways for opencode and pi +├── subagents.nix # frontmatter parser + per-agent renderers └── ponytail.nix +rules/ # code-style, subagents (new) +subagents/ # 10 agents: 5 rewritten, 5 new +providers/ # gateway model discovery scripts ``` How each agent reads a secret (`env` or `headers` value `{ file = ...; }`) at startup: | Agent | stdio `env` | HTTP `headers` | | --- | --- | --- | | Claude Code | wrapper script | `headersHelper` | | codex | wrapper script | `env_http_headers` + codex wrapper | | opencode | `{file:...}` | `{file:...}` | | pi | `!cat` | `!cat` | ## Evidence - **Before:** `nix flake check` covered skill frontmatter and lint only; the module was never evaluated. **After:** new `subagents` check (parses and renders all 10) and `module` check (full Home Manager evaluation, every part on, secret-backed stdio + HTTP MCP servers) -> `all checks passed!` - Built and inspected the generated pieces: the Claude `headersHelper` emits `{"X-Api-Key": "<file contents>"}`; the codex wrapper contains `export MCP_HTTP_SECRET_X_API_KEY="$(cat /run/secrets/key)"`; read-only subagents render with `sandbox_mode = "read-only"` (codex), `disallowedTools` (Claude), `tools: {...: false}` (opencode), a reduced allowlist (pi). - Not done: a real switch on a host, or launching each agent against the generated config. ## Merge Danger **Door:** two-way Nothing consumes this until the NixOS config is migrated; reverting the merge restores `m4rti-skills`. **Blast Radius:** consumers `m4rti-skills.*` is removed, so the NixOS config fails to evaluate on the next `nix flake update skills` until it moves to `m4rti.agents`. When migrating: drop the `CODEX_HOME` session variable (it would hide the Home Manager codex config), drop the separate zeit module import (this module imports it), and note Claude Code now defaults to the llm-agents build on every host. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

marti added 4 commits 2026-10-13 19:38:01 +02:00

feat: add tool-neutral subagents and their per-agent renderer ... 8f7d173c8d

Rewrites the five opencode-only subagents from the NixOS config into a
single-line frontmatter format (name, description, deny, mode) that
nix/subagents.nix parses and renders for Claude Code, opencode, codex and
pi. Adds explorer, code-reviewer, test-writer, debugger and docs-writer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

feat: add global rules and LiteLLM gateway discovery scripts ... 559fddb70e

Moved from the NixOS config. The time-tracking rule is left behind since
the zeit plugin tracks turns itself; a new rule tells agents to delegate
long or wide tasks to subagents.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

feat!: replace m4rti-skills with the m4rti.agents module ... e23267fa71

One Home Manager module for the whole agent environment: apps (Claude
Code and Desktop, codex, ChatGPT desktop, opencode CLI and desktop, pi,
T3 Code), skills, rules, MCP servers with file-backed secrets, plugins
(ponytail, graft, open-code-review, zeit, agent-browser), subagents and
model gateways. Adds a check that evaluates a full configuration.

BREAKING CHANGE: `m4rti-skills.*` is gone; use `m4rti.agents.*`.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs: document the agent-env module ... 3adaa7e6f2

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

marti merged commit 3ca27fd4d3 into main 2026-10-13 19:42:03 +02:00

marti referenced this pull request from a commit 2026-10-13 19:42:03 +02:00

Merge pull request 'feat!: turn the skills repo into agent-env (m4rti.agents module)' (#2) from feature/agent-env into main

marti deleted branch feature/agent-env 2026-10-13 19:42:05 +02:00

marti referenced this pull request 2026-10-13 19:47:03 +02:00

feat!: pin llm-agents to its own nixpkgs, move codex to apps.gpt.codex #3

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

marti/agent-env!2

No description provided.