Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 22 additions & 10 deletions skills-stub/README.md → .agents/skills/README.md
Original file line number Diff line number Diff line change
@@ -1,15 +1,17 @@
# Bezel skills-stub
# Bezel skills

This directory is the skills input for the Bezel flake until you point
`inputs.skills` at your own skills repo. The flake uses this path:
This directory is the canonical skills tree and the flake input:

```nix
inputs.skills.url = "path:./skills-stub";
inputs.skills.url = "path:./.agents/skills";
```

When you have a skills repo, change only that URL (and lock). Do **not** add
a second repo-root `skills/` tree — the Home Manager module only symlinks
`${inputs.skills}`.
a repo-root `skills/` tree, `.omp/skills/`, or `.codex/skills/`.

`scripts/sync-skills` rewrites the regular-file copies in `.claude/skills/`,
`.grok/skills/`, and `.cursor/skills/` from this tree. Edit here, then sync.
Those copies are not a second source.

Shipped skills:

Expand All @@ -28,8 +30,18 @@ prints the same JSON envelope: `{ ok, command, exit, data, error }`. The
Nushell that runs them is the one pinned by this flake's nixpkgs lock. MCP
servers for those tools stay supported; this tree does not turn them off.

Copy or link one directory into another harness's skills folder
(`~/.agents/skills/` or `~/.claude/skills/`). With Home Manager, the whole
input is already linked at `~/.config/omp/agent/skills`.
`programs.bezel.enable` copies each skill directory, as regular files, into:

- `~/.agents/skills`
- `~/.claude/skills`
- `~/.grok/skills`
- `~/.cursor/skills`

A second run leaves a file alone when its bytes already match, and it keeps
that file's mode. If a managed file already exists with different bytes, or
the path is a symlink, `scripts/install-user-skills.sh` prints a drift report
and writes nothing. `--force` replaces the bytes and keeps the existing mode.
`--check` reports drift and writes nothing. Unrelated skill directories
already in those folders are left in place.

See [docs/SPIKE-stanley.md](../docs/SPIKE-stanley.md).
See [docs/SPIKE-stanley.md](../../docs/SPIKE-stanley.md).
13 changes: 13 additions & 0 deletions .agents/skills/bootstrap/SKILL.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,13 @@
---
name: bootstrap
description: Placeholder skill shipped with the Bezel skills tree until a skills repository is published. Replace the flake skills input. Harness copies are generated from .agents/skills.
---

# bootstrap

This skill exists so the skills flake input is a valid skills directory
(`*/SKILL.md`) until you point the flake skills input at a skills repository.

Home Manager copies each skill directory as regular files into the user
directories the harnesses read: `~/.agents/skills`, `~/.claude/skills`,
`~/.grok/skills`, and `~/.cursor/skills`.
File renamed without changes.
File renamed without changes.
File renamed without changes.
File renamed without changes.
File renamed without changes.
File renamed without changes.
File renamed without changes.
File renamed without changes.
File renamed without changes.
File renamed without changes.
13 changes: 13 additions & 0 deletions .claude/skills/bootstrap/SKILL.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,13 @@
---
name: bootstrap
description: Placeholder skill shipped with the Bezel skills tree until a skills repository is published. Replace the flake skills input. Harness copies are generated from .agents/skills.
---

# bootstrap

This skill exists so the skills flake input is a valid skills directory
(`*/SKILL.md`) until you point the flake skills input at a skills repository.

Home Manager copies each skill directory as regular files into the user
directories the harnesses read: `~/.agents/skills`, `~/.claude/skills`,
`~/.grok/skills`, and `~/.cursor/skills`.
50 changes: 50 additions & 0 deletions .claude/skills/check/SKILL.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,50 @@
---
name: check
description: Bounded Jev-first check of a git diff. Code gathers evidence, Jev classifies, code applies thresholds. Empty findings are not approval.
---

# check

Thin overlay skill for a bounded diff check. Prefer this
workflow when the user asks to check a diff against a task.

## Run (shadow default)

```sh
# Key from the environment only — never flags or files.
export TYPESAFE_API_KEY # host card store / your shell; omit to smoke without live Jev

# Worktree + staged, or inject a unified diff:
jev-router --check --intent "<task or request>"
jev-router --check --intent "<task>" --base origin/main
jev-router --check --intent "<task>" --diff-file path/to.diff
```

`JEV_MODE` defaults to `shadow`: Choice is logged; **shadow never blocks**.
Missing `TYPESAFE_API_KEY` in shadow continues — hunks land in `notChecked`.

## Result contract

Stdout is one JSON object:

| Field | Meaning |
| --- | --- |
| `findings` | Deterministic or Jev-threshold hits |
| `parked` | Unclear; a human should look |
| `notChecked` | Skipped, unjudged, or out of scope |
| `approval` | Always `false` on this path |
| `emptyFindingsAreNotApproval` | Always `true` |

**Empty `findings` is not approval.** Do not tell the user the change is
approved, LGTM, or safe to merge because the list is empty.

## Available-gate

`check` (and the `review` routing stub) require a git diff. No diff →
`status: "no_diff"` and `cannot_tell` — do not invent a pass.

## Do not

- Invoke a `stanley` binary, another agent binary, improve-worker, or promote loop
- Pass `TYPESAFE_API_KEY` as a flag or write it to a file
- Auto-activate agent-drafted skills
44 changes: 44 additions & 0 deletions .claude/skills/facet/SKILL.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,44 @@
---
name: facet
description: >
Inspect Facet HTTP request history, sessions, pins, and stored-value
metadata through the facet CLI. Use when a task needs to know what requests
ran, their status and timing, or to compare two runs. Prefer this skill over
the facet MCP server: it adds no tool schemas to context. The MCP server
remains available.
---

# facet — request history skill

Read-only. The body is the fixed script `facet.nu` next to this file. Every
wrapped command appends `--json` (facet expects it after the subcommand).
Do not reimplement the calls.

## Commands

```bash
nu facet.nu history [path] [--limit 20] [--request sel] [--status 200] [--session id]
nu facet.nu last [path] [--request sel]
nu facet.nu sessions [--open]
nu facet.nu session <id|current>
nu facet.nu pins [path]
nu facet.nu env [path] [--environment e] # names only, never values
nu facet.nu diff <idA> <idB> [path] # hashes only
nu facet.nu doctor
```

Each command prints one JSON envelope: `{ ok, command, exit, data, error }`.
`diff` uses `exit` 1 to mean the runs differ (`data.same` is false); that is
still `ok: true`.

## Not wrapped, on purpose

`facet.nu` refuses these and does not call facet:

- `history --bodies`, `history --id`, `blob`, `diff --bodies`: these can print
recorded request or response bodies and headers, which may carry tokens.
- `request run`, including `request run --dry-run`: a preview resolves
variables, so it can show a hydrated secret.
- `replay`, `env set`, `env delete`, `pin`, `gc --yes`,
`session start/end`: these change state or send traffic. Run them only when
a human has explicitly asked, and never pass a secret value on the command line.
Loading
Loading