Bullish-Design

Use for practical command recipes when agents operate Allium CLI commands on a folder of .allium specs.

Use when running deeper semantic and process-level analysis on a folder of .allium specs after allium check has passed.

Use when validating a folder of .allium specs with allium check. Resolves the folder to an entrypoint file, runs validation, interprets diagnostics, and defines the repair loop.

Use when defining or applying CI-quality command gates for a folder of .allium specs. Assumes allium is already available.

Use as a concise reference for Allium CLI commands when working on a folder of .allium specs. Assumes the CLI is already available and entrypoint resolution is required.

Use when an Allium CLI command fails on a folder of .allium specs. Classifies diagnostics, performs minimal repairs, and reruns the correct command against the resolved entrypoint.

Use when extracting or inspecting the structured domain model from a folder of .allium specs using allium model.

Use when summarizing, comparing, or acting on Allium CLI output from commands run against a folder of .allium specs.

Use when writing a scripts/tasks/processes entry or a manager module. Covers which surface to use, guarding enterShell echoes, and the 0/1/2/3 exit contract.

Use when editing devenv.yaml or adding an input/import. Covers flake:false for modules, nixpkgs-python for version pins, and transitive inputs.

Use when a module edit or an input bump doesn't take effect, or to decide which cache layer to reset. The lock vs eval-cache decision tree for devenv repos.

Use when you edited a modules/ file, devenv.nix, or env.* and nothing changed. Resolves the lock + eval-cache staleness loop.

Use when starting a server or any long-running / heavy task. Use processes + devenv up; don't block the shell; poll logs instead of piping out of view.

Use on ModuleNotFoundError or import failures inside the devenv shell, or after a fresh venv. The venv exists but deps may not be installed.

Use when about to run pytest, python, uv, ruff, or any build/tooling command in this repo. Enforces running through the devenv shell.

Use on "command not found", the shell won't enter, or eval errors. A symptom→cause→fix decision tree for devenv repos.

Orientation and mental model for the Cairn runtime: the canonical working tree, the four runtime layers (code providers, repository snapshots, bubblewrap execution, orchestration), the data layout, the agent lifecycle state machine, and the canonical docs.…

Operating a Cairn deployment from the command line: starting the daemon (cairn up) or running inline (cairn run), submitting work (spawn/queue), the review flow (list-agents, status, accept with staleness checks, reject, undo, logs), the Unix-socket transport…

Extending Cairn with pluggable code sources: the CodeProvider protocol, the built-in file and inline providers, plugin entry-point registration under the cairn.providers group, the provider context dict, and validation semantics. Use when adding a custom code…

Contributing to the Cairn codebase itself: the devenv development gate (devenv test), test layout and conventions, the SPEC.md source-of-truth rule, coding conventions, the security model, performance targets, and common pitfalls when modifying src/cairn,…

Embedding Cairn as a library: open_workspace and WorkspaceManager (WAL/MVCC, readonly, busy_timeout, max_content_bytes), WorkspaceInspector and WorkspaceStats, AgentStateManager (namespaced KV with typed models and atomic counters), TaskQueue, retry…

How to write Python task scripts that run inside Cairn's bubblewrap sandbox: the nine injected sandbox helpers (read_file, write_file, list_dir, file_exists, delete_file, search_files, search_content, submit_result, log), path rules, resource limits, the…

Route ALL version control through gitman (jj + colocated git). Never run raw jj/git.

Adopt or templatize an existing repo — bring a hand-written repo under template management with copyroom adopt / templatize, and read the drift report.

Entry point for all CopyRoom work — scaffolding a repo from a template, template drift and updates, and adopting an existing repo. Routes to the copyroom-adopt and copyroom-template-edit sub-skills for the deep arcs.

Edit a template from inside a generated project — the template-checkout → edit → template-test → template-preview → template-discard loop, preview-only.

The user's standing agent configuration — the cross-repo law that holds in every one of their repositories. Read it alongside this repo's AGENTS.md; AGENTS.md says what this project is, this skill says how the user wants work done anywhere.

Develop the pydantree library itself (pydantree_sitter / pydantree_sitter_grammar). Use when editing library code, running the test suite, debugging builds or scanners, or making changes that touch the package layout, pipeline, loader, or schema. Includes the…

Extract typed data from text with pydantree_sitter (Product A) — OutputModel declarations, captures (field/kind/record/optional/list/descendant), predicates and markers, schema binding and validate_with, bundles, community grammars, stubs, and the error…

Author tree-sitter grammars with pydantree_sitter_grammar (Product B) — the Pydantic DSL, static checks, the conflict loop, expressions/precedence ladders, external scanners, the corpus harness, bundles, and the community-grammar schema tool. Use when…

The pydantree external-scanner library — the airtight mechanism (externals in scanner order, scanner= to the build, content-addressed cache, ExternalScannerRequiredError), the two gotchas (mid-whitespace scans; multiple externals valid in one state), the five…

Use on ModuleNotFoundError or import failures inside the devenv shell, or after a fresh venv. The venv exists but deps may not be installed.

Use on "command not found", the shell won't enter, or eval errors. A symptom→cause→fix decision tree for devenv repos.

Route ALL version control through gitman (jj + colocated git). Never run raw jj/git.

Start here for any work in this repo. Routes to the right manager and owns the lifecycle order (verify before save, scaffold before change).

Bring a repository into the devman automation plane — the two devenv.yaml lines, the three devenv.nix keys, the task names its groups call, and the proof. Holds the per-ecosystem task recipes and the adoption failures measured across the rollout.

Entry point for all devman work — the Dagu automation plane. Use it to run, inspect and diagnose workflows in any registered repository, and to route to the devman-workflow and devman-adopt sub-skills for authoring a workflow or joining a repository to the…

Author or change a devman workflow — a repository's own .devman/workflows/ file, a group workflow, or a triggers.toml mapping. Holds the decision tree, the file rules, the parameter and queue conventions, the loop-break and schedule mechanics, and the test…

Route ALL version control through gitman (jj + colocated git). Never run raw jj/git.

Use when writing or editing a Typer CLI in this repo — declaring commands, arguments vs options, the 0/1/2/3 exit-code contract via typer.Exit, no_args_is_help, and testing with CliRunner. Curated for typer 0.12.

Use when verifying code changes in this repository — running tests, lint, formatting, or type checks, or checking whether a change is correct. This repo verifies through Testee, not by invoking pytest/ruff/ty directly.

An LLM-native language for sharpening intent alongside implementation. Velocity through clarity.

Generate tests from Allium specifications. Use when the user wants to propagate tests, generate test files from a spec, write tests for a specification, create property-based tests, produce state machine tests, check test coverage against spec obligations, or…

Use when you need to understand what projects exist in this workspace and how they integrate — or execute one command across selected projects — before exploring repos by hand. This workspace is mapped by fleetman; read its generated index instead of…

Route ALL version control through gitman (jj + colocated git). Never run raw jj/git.

Author or edit the scripted tyo3.nvim demo recordings (vhs .tape → .gif/.txt/.cast). Use when creating a new demo scene, editing an existing tour/context tape, changing the demo project or nvim init, fixing a racey/broken recording, adding a new demo variant,…

Review a tyo3.nvim demo recording by reading its .txt transcript frame-by-frame. Use when verifying a re-rendered demo, checking a recording landed correctly, diagnosing a broken/racey/half-typed GIF, confirming a scene shows the right…

Locate and extract authoritative documentation from this repository's local context/vendor materials when they exist.

Template workflow for diagnosing LSP chat-delivery regressions when pyagent introduces an LSP/runtime pipeline.