commitizen-tools
diff --git a/AGENTS.md b/AGENTS.md
@@ -2,82 +2,61 @@
 
 ## Purpose
 
-This file provides **project-specific guidance for AI agents** (and other automated tools) working on the `commitizen` repository.
-Follow these instructions in addition to any higher-level system or tool rules.
+This file is the auto-loaded entry point for AI agents working on the `commitizen` repository. It holds the rules an agent needs in **every** session. Deeper guidance lives in:
 
-## Project Overview
+- [Contributing](docs/contributing/contributing.md) — setup, dev workflow, PR lifecycle.
+- [Contributing TL;DR](docs/contributing/contributing_tldr.md) — poe command cheat sheet.
+- [Pull Request Guidelines](docs/contributing/pull_request.md) — PR etiquette and AI-assisted policy.
+- [Architecture Overview](docs/contributing/architecture.md) — codebase topology and extension points.
+- [For AI Agents](docs/contributing/agents/index.md) — agent-shaped recipes, validation map, playbooks.
 
-- **Project**: `commitizen` - a tool to help enforce and automate conventional commits, version bumps, and changelog generation.
-- **Primary language**: Python (library + CLI).
-- **Cross-platform**: Tests run on Linux, macOS, and Windows. Avoid POSIX-only assumptions in code (paths, subprocesses, line endings).
-- **Key entrypoints**:
-  - `commitizen/cli.py` - main CLI implementation.
-  - `commitizen/commands/` - subcommands such as `bump`, `commit`, `changelog`, `check`, etc.
-  - `commitizen/config/` - configuration discovery and loading.
-  - `commitizen/providers/` - version providers (e.g., `pep621`, `poetry`, `npm`, `uv`).
-- **Config sources**: `pyproject.toml` (project config, poe tasks, ruff, mypy), `.pre-commit-config.yaml` (hooks), `.github/workflows/` (CI).
-
-## General Expectations
-
-- **Preserve public behavior and CLI UX** — no breaking changes to APIs, CLI flags, or exit codes unless explicitly requested.
-- **Update or add tests/docs** when you change user-facing behavior.
-- **Commit messages** must follow [Conventional Commits](https://www.conventionalcommits.org/) (enforced by commitizen itself).
-- **Pull requests** must follow the [Pull Request Guidelines](docs/contributing/pull_request.md) and the template in `.github/pull_request_template.md`.
-
-## Setup and Validation
-
-> Full contributor guidelines (prerequisites, workflow, PR process): [`docs/contributing/contributing.md`](docs/contributing/contributing.md).
-
-### Bootstrap
-
-```bash
-uv sync --frozen --group base --group test --group linters
-uv run poe setup-pre-commit   # install git hooks (uses prek, a pre-commit runner)
-```
-
-### Local commands
-
-- **Format**: `uv run poe format` (runs `ruff check --fix` then `ruff format`)
-- **Lint**: `uv run poe lint` (runs `ruff check` then `mypy`)
-- **Test**: `uv run poe test` (runs `pytest -n auto`)
-- **CI-equivalent**: `uv run poe ci` (commit check + pre-commit hooks via `prek` + test with coverage)
-- **Full local check**: `uv run poe all` (format + lint + check-commit + coverage)
+Follow these instructions in addition to any higher-level tool or system rules.
 
-Always run at least `uv run ruff check --fix . && uv run ruff format .` before pushing. CI will fail if the formatter modifies any files.
+## Project at a glance
 
-### CI pipeline
-
-- CI runs `poe ci` on a matrix of Python 3.10–3.14 × ubuntu/macos/windows.
-- Pre-commit hooks are defined in `.pre-commit-config.yaml` and run via [`prek`](https://github.com/j178/prek) (a `pre-commit` compatible runner).
-- The matrix is **fail-fast**: inspect the earliest failing job that completed; others are cancelled.
-
-### Common CI failure patterns
-
-- **"Format Python code...Failed"**: Run `uv run poe format` and commit the result.
-- **mypy `[arg-type]` on TypedDict**: Dynamically-constructed dicts (e.g., from `pytest.mark.parametrize`) passed to TypedDict-typed params need `# type: ignore[arg-type]`.
-- **"pathspec 'vX.Y.Z' did not match"**: `.pre-commit-config.yaml` pins a tag of this repo. Rebase onto master to pick up the tag.
-- **`VersionProtocol` + `issubclass`**: This Protocol has non-method members (properties), so `issubclass()` raises `TypeError`. Use `hasattr` checks for runtime validation.
+- **Project**: `commitizen` — Python CLI for enforcing Conventional Commits, automating version bumps, and generating changelogs.
+- **Library + CLI**: code is reachable both via `cz` and `import commitizen`.
+- **Cross-platform**: tests run on Linux/macOS/Windows × Python 3.10–3.14. Avoid POSIX-only assumptions (paths, subprocesses, line endings).
+- **Key entrypoints**:
+    - `commitizen/cli.py` — CLI definition (decli + argparse).
+    - `commitizen/commands/` — one module per `cz` subcommand.
+    - `commitizen/config/` — configuration discovery and parsing.
+    - `commitizen/providers/` — version providers.
+    - `commitizen/changelog_formats/` — changelog file formats.
 
-## What to Read Before Changing
+## Read before changing
 
 | Changing... | Read first |
 |---|---|
 | CLI flags/arguments | `commitizen/cli.py`, `docs/commands/<cmd>.md`, `tests/test_cli/` |
 | Bump logic | `commitizen/bump.py`, `commitizen/commands/bump.py`, `docs/commands/bump.md` |
 | Changelog generation | `commitizen/changelog.py`, `commitizen/changelog_formats/`, `docs/commands/changelog.md` |
 | Version schemes | `commitizen/version_schemes.py`, `tests/test_version_schemes.py` |
-| Version providers | `commitizen/providers/`, `tests/test_providers.py`, `docs/config/version_provider.md` |
+| Version providers | `commitizen/providers/`, `tests/providers/`, `docs/config/version_provider.md` |
 | Config resolution | `commitizen/config/`, `tests/test_conf.py`, `docs/config/` |
 | Tag handling | `commitizen/tags.py`, `tests/test_tags.py` |
 | Pre-commit / CI | `.pre-commit-config.yaml`, `.github/workflows/`, `pyproject.toml` (poe tasks) |
 
-## Coding Guidelines
+For recurring task types (add a provider, deprecate an API, regenerate snapshots, ...), use the matching playbook in [For AI Agents § Playbooks](docs/contributing/agents/index.md#playbooks) instead of reinventing the workflow.
+
+## Do not touch
+
+These files are regenerated by Commitizen-specific tooling, so hand-edits get reverted on the next release or doc rebuild:
+
+- `CHANGELOG.md` — produced by `cz changelog`. Hand-edits will be overwritten on the next release.
+- `commitizen/__version__.py` — bumped by `cz bump` via the configured version provider.
+- `.pre-commit-config.yaml:rev:` lines under the `Commitizen` repo — bumped by `cz bump` (`version_files` in `pyproject.toml`).
+- `docs/images/cli_help/*.svg` and `docs/images/cli_interactive/*.gif` — regenerated by `uv run poe doc:screenshots`. See the [update-snapshots playbook](docs/contributing/agents/playbooks/update-snapshots.md).
+- `tests/**/*` snapshot files used by `pytest-regressions` — regenerated via `uv run poe test:regen`.
+
+## Mandatory PR reminders
+
+These are easy to miss when working from an agent and are required by the PR template:
 
-- **Types**: Preserve or improve existing type hints.
-- **Errors**: Prefer `commitizen/exceptions.py` error types; keep messages clear for CLI users.
-- **Output**: Use `commitizen/out.py`; do not add noisy logging.
+1. **Complete the AI disclosure**. Check "Was generative AI tooling used to co-author this PR?" and fill in the `Generated-by:` trailer with the tool name. Details: [Pull Request Guidelines § AI-Assisted Contributions](docs/contributing/pull_request.md#ai-assisted-contributions).
+2. **Run `uv run poe all` before pushing**. This is the command named in the PR template; it auto-formats then runs the same lint/check/test pipeline as CI. To mirror CI exactly afterwards, run `uv run poe ci` (uses `prek`, does not auto-format).
+3. **Fill in "Steps to Test This Pull Request"** with the exact commands you ran locally — the maintainers re-run them.
 
-## When Unsure
+## When unsure
 
-- Prefer **reading tests and documentation first** to understand the expected behavior.
-- When behavior is ambiguous, **assume backward compatibility** with current tests and docs is required.
+When behavior is ambiguous, assume **backward compatibility with current tests and docs** is required. Add a deprecation window instead of breaking it; see the [deprecate-public-api playbook](docs/contributing/agents/playbooks/deprecate-public-api.md).
diff --git a/docs/contributing/agents/index.md b/docs/contributing/agents/index.md
@@ -0,0 +1,38 @@
+# For AI Agents
+
+These pages are written for AI agents contributing to Commitizen. Human contributors may also find them useful as a quick reference. They **complement** the existing human-facing contributor docs rather than replace them — anything covered by the human docs is linked, not restated.
+
+> If you are an AI agent looking to **use** Commitizen as a tool (validate
+> commit messages, bump versions in a downstream project), see the skill
+> definition at `.agents/skills/commitizen/SKILL.md` in the repo root.
+> The pages here are for working **on** Commitizen itself.
+
+## When to read what
+
+| You want to... | Read |
+|---|---|
+| Set up a local dev environment | [Contributing](../contributing.md#prerequisites-setup) |
+| Look up a poe command | [Contributing TL;DR](../contributing_tldr.md#command-cheat-sheet) |
+| Understand the codebase layout and extension points | [Architecture Overview](../architecture.md) |
+| Open a pull request | [Pull Request Guidelines](../pull_request.md) and the [PR template](https://github.com/commitizen-tools/commitizen/blob/master/.github/pull_request_template.md) |
+| Pick the right test selector for a change | [Validation Guide](validation.md#targeted-test-map) |
+| Recover from a CI failure | [Validation Guide](validation.md#ci-failure-recipes) |
+| Implement a recurring task type | [Playbooks](#playbooks) |
+
+The repo-root [`AGENTS.md`](https://github.com/commitizen-tools/commitizen/blob/master/AGENTS.md) is the auto-loaded entry point for most agent tools. It holds the rules an agent needs in every session; this page is the deeper reference.
+
+## Playbooks
+
+Recipes for recurring task types. Each playbook is self-contained: trigger, files to read first, ordered steps, verification commands, and known pitfalls. They link out to the human-facing concept docs rather than restating concepts.
+
+- [Add a version provider](playbooks/add-version-provider.md)
+- [Add a changelog format](playbooks/add-changelog-format.md)
+- [Add or modify a CLI flag](playbooks/add-cli-flag.md)
+- [Deprecate a public API](playbooks/deprecate-public-api.md)
+- [Update generated snapshots and screenshots](playbooks/update-snapshots.md)
+
+If no playbook matches, read the [Architecture Overview](../architecture.md) for the relevant subsystem and follow 1-2 existing examples in the same directory before changing code.
+
+## Updating these pages
+
+Treat these pages like any other code change: open a PR, follow the template, run `uv run poe doc:build` to verify the mkdocs build, and check internal links. If you find yourself restating something that already lives in a human-facing doc, link to it instead and shorten the agent doc.
diff --git a/docs/contributing/agents/playbooks/add-changelog-format.md b/docs/contributing/agents/playbooks/add-changelog-format.md
@@ -0,0 +1,69 @@
+# Playbook: Add a Changelog Format
+
+A changelog format handles parsing and rendering a `CHANGELOG.<ext>` file in a specific markup language. End-user documentation: [Changelog command](../../../commands/changelog.md). Built-ins are `markdown` (default), `asciidoc`, `textile`, `restructuredtext`.
+
+Architectural context: [Architecture § Extension points](../../architecture.md#extension-points).
+
+## Trigger
+
+- "Support `<markup>` changelogs."
+- A user wants `cz changelog` to emit something other than Markdown.
+- An incremental-changelog use case fails because the user's existing `CHANGELOG` file is not Markdown.
+
+## Read first
+
+- `commitizen/changelog_formats/__init__.py` — `ChangelogFormat` Protocol, entry-point group `commitizen.changelog_format`, `KNOWN_CHANGELOG_FORMATS` registry, `_guess_changelog_format` extension-based fallback.
+- `commitizen/changelog_formats/base.py:BaseFormat` — provides a default `get_metadata_from_file` that walks the file once for **line-anchored** titles. You override `parse_version_from_title` and `parse_title_level` to fit your markup.
+- A close-match existing format:
+    - Line-anchored heading prefix (`#`, `=`, `h1.`): `commitizen/changelog_formats/markdown.py`, `asciidoc.py`, or `textile.py` — each overrides only the two `parse_*` methods.
+    - Multi-line titles (overline/underline, as in reStructuredText): `commitizen/changelog_formats/restructuredtext.py` — overrides `get_metadata_from_file` directly because the base class assumes one-line titles.
+- `commitizen/templates/` — Jinja2 templates named `CHANGELOG.<ext>.j2` (e.g. `CHANGELOG.md.j2`, `CHANGELOG.rst.j2`, `CHANGELOG.adoc.j2`); the file extension comes from the `extension` class attribute, not the format name.
+- `tests/test_changelog_format_<name>.py` — every format has parity tests; copy the closest one.
+
+## Steps
+
+1. **Create the format module** at `commitizen/changelog_formats/<name>.py`. Subclass `BaseFormat`. Set the class attributes:
+    - `extension: ClassVar[str]` — primary file extension (no dot).
+    - `alternative_extensions: ClassVar[set[str]]` — other accepted extensions for the same format.
+2. **Implement two methods** for line-anchored formats:
+    - `parse_version_from_title(line: str) -> VersionTag | None` — given one line, return a `VersionTag` if the line is a release heading.
+    - `parse_title_level(line: str) -> int | None` — return the heading level (1, 2, 3, ...) if the line is a heading. The base class `BaseFormat.get_metadata_from_file` walks the file once and calls both methods per line.
+
+    If your format's titles span multiple lines (overline/title/underline, as in reStructuredText), override `get_metadata_from_file` directly instead — see `RestructuredText` for the algorithm.
+3. **Add the Jinja2 template** at `commitizen/templates/CHANGELOG.<ext>.j2`. Mirror the structure of `CHANGELOG.md.j2` — same blocks, different markup. Make sure the loops over `tree`, `entries`, and `change_type` match.
+4. **Register the built-in** in `pyproject.toml` under `[project.entry-points."commitizen.changelog_format"]`:
+
+   ```toml
+   <name> = "commitizen.changelog_formats.<name>:<Name>"
+   ```
+
+5. **Add tests** at `tests/test_changelog_format_<name>.py`. Copy the closest existing test file and adapt the fixtures.
+6. **Update the cross-format suite** `tests/test_changelog_formats.py` if it parametrizes over all formats — add the new one to its lists.
+7. **Update user docs** at `docs/commands/changelog.md` and `docs/customization/changelog_template.md` — list the new format and show how to opt in via `changelog_format`.
+8. **Re-run the install** so the entry point registers:
+
+   ```bash
+   uv sync --frozen --group base --group test --group linters
+   ```
+
+## Validate
+
+```bash
+uv run pytest tests/test_changelog_format_<name>.py tests/test_changelog_formats.py tests/test_changelog.py tests/test_incremental_build.py -n auto
+uv run poe lint
+uv run poe doc:build      # if docs changed
+uv run poe all            # final pre-push
+```
+
+## Pitfalls
+
+- **`KNOWN_CHANGELOG_FORMATS` is populated at import time** from entry points, so you must re-run `uv sync` after editing `pyproject.toml` before tests can see your new format.
+- **Forgetting `alternative_extensions`** — `_guess_changelog_format` uses both `extension` and `alternative_extensions` when the user does not set `changelog_format` explicitly. If a user has `CHANGELOG.<alt-ext>`, your format will not auto-detect without it.
+- **Template encoding** — Jinja2 reads templates with the active encoding; keep them ASCII-safe or test with non-UTF-8 `encoding` settings.
+- **Heading regex anchoring** — match the whole line (`^...$`) when the markup is line-anchored (Markdown headings); a substring match will pick up non-heading lines that mention `unreleased`.
+- **Snapshot updates** — many changelog tests use `pytest-regressions`. See the [update-snapshots playbook](update-snapshots.md) when output intentionally changes.
+
+## Stop and ask if
+
+- The target format requires structured metadata that does not fit the `parse_title_*` Protocol (e.g., front-matter in YAML).
+- The format implies a fundamentally different rendering tree (e.g., one file per release) — that is a bigger change than a format addition.