git:20261108.f69b1b1 to git:20261108.6e88f22

# Repository Guidelines

## Project Structure & Module Organization

- `cmd/mistermorph/`: CLI entrypoint and subcommands.

- `agent/`: agent engine (loop, parsing, prompts, logging).

- `llm/`: shared LLM types/interfaces used by the agent and providers.

- `providers/`: LLM backends (currently `providers/uniai/`).

- `tools/` and `tools/builtin/`: tool registry and built-ins (`web_search`, `url_fetch`, `bash`, `read_file`).

- `skills/`: skill discovery and selection logic for `SKILL.md`.

- `demo/`: embedding examples (`demo/embed-go/`, `demo/embed-cli/`).

- `wrappers/acp/`: optional stdio bridges for external ACP agents (Codex app-server, Claude Code print mode, Cursor CLI `agent acp` proxy).

- Root configs: `assets/config/config.example.yaml` (template) and `config.yaml` (local). `mistermorph` in repo root is a build artifact.

## Build, Test, and Development Commands

- Build: `go build -o ./bin/mistermorph ./cmd/mistermorph`

- Run (no build): `go run ./cmd/mistermorph --help`

- Test: `go test ./...`

- Static checks: `go vet ./...`

- Example run:

- `./bin/mistermorph run --task "Summarize this repo" --provider openai --model gpt-5.2 --api-key "$OPENAI_API_KEY"`

## Demos (Embedding)

- Go library demo: `cd demo/embed-go && OPENAI_API_KEY="..." go run . --task "List files and summarize."`

- CLI subprocess demo: `go build -o ./bin/mistermorph ./cmd/mistermorph`, then `cd demo/embed-cli && MISTER_MORPH_BIN=../../bin/mistermorph OPENAI_API_KEY="..." go run . --task "Search for OpenAI and fetch the first result"`

- Both demos require network access (OpenAI-compatible provider).

## Coding Style & Naming Conventions

- Format Go code with `gofmt` (`go fmt ./...` is fine for day-to-day).

- Package names are lowercase; exported identifiers use `PascalCase`; locals use `camelCase`.

- Keep provider-specific behavior in `providers/<name>/` and avoid leaking it into `agent/`.

## Testing Guidelines

- Use the standard `testing` package; colocate tests as `*_test.go` next to the code they cover.

- Prefer table-driven tests and subtests (`t.Run("case", ...)`) for tool/provider edge cases.

## Agent Hooks & Webhooks

- There’s no built-in inbound webhook/HTTP server.

- The engine supports step hooks via `agent.Hook` + `agent.WithHook(...)` (runs once per step, before the LLM call).

- For outgoing webhooks and complex HTTP needs, implement it in a hook or use a tool; built-in `url_fetch` supports GET/POST/PUT/DELETE with optional headers/body.

## Security & Configuration Tips

- Prefer `MISTER_MORPH_API_KEY` over committing `api_key` in config; treat logs as sensitive when enabling debug/thought output.

- When adding/changing config keys, update `assets/config/config.example.yaml` (the template used for docs and examples).

## Model Catalogs

- Model pricing comes from uniai (`uniai.DefaultPricingCatalog`, picked up by bumping the uniai version). Each time model pricing is updated (new models or new prices), update the context window sizes in `llm/model_context_windows.yaml` as well: add every new model (with its common aliases), and correct changed windows.

- Take window sizes only from the provider's official docs, and record the page in `sources` with the `checked_at` date. Do not guess a size; leave a model out until its docs state one.

- Add new models to the lookup checks in `llm/model_context_window_catalog_test.go`.

## Commit & Pull Request Guidelines

- This checkout may not include Git history; if it does, match existing conventions. Otherwise, use Conventional Commits (`feat:`, `fix:`, `docs:`, `refactor:`).

- PRs should include: a clear description of the behavior change, how to run it locally, and any config/flag updates.