# Tcr Kentbeck

> >

- **Type:** Skill
- **Install:** `agentstack add skill-xpepper-tcr-skill-tcr-kentbeck`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [xpepper](https://agentstack.voostack.com/s/xpepper)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [xpepper](https://github.com/xpepper)
- **Source:** https://github.com/xpepper/tcr-skill/tree/main/tcr-kentbeck

## Install

```sh
agentstack add skill-xpepper-tcr-skill-tcr-kentbeck
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# TCR (Test && Commit || Revert)

## Non-negotiables

When this skill applies, treat TCR as **mandatory**, not optional:

1. **One small step** — Prefer the smallest change that could move tests toward green (or add a failing test first in strict TDD, then minimal implementation).
2. **Run tests** — Use the project’s real test command (see [Detecting the test command](#detecting-the-test-command)). **Always run the full suite** for that command (same as CI / default project test entrypoint): e.g. `pytest` with no path arguments, `npm test` / `pnpm test` / `yarn test`, `cargo test`, `go test ./...`, `make test` if that runs everything. Do **not** narrow to a single file, class, or `-k` filter unless the user explicitly asks for a scoped run.
3. **Green → commit** — If tests pass, **commit** that step immediately with a message that describes *what* changed (present tense, imperative).
4. **Red → revert** — If tests fail, **do not** keep the change. Restore the working tree to the last committed state (`git restore .` / `git checkout -- .` or revert unstaged edits as appropriate). Then retry with a smaller or different step.

5. **Red → log** — After reverting, **append** an entry to **`tcr-failure-log.md`** at the repository root (create it with a short title if missing). Record what went wrong so the failure is not invisible: **when** (ISO timestamp or equivalent), **which test(s)** failed and the **assertion or error** (or a short traceback excerpt), **what change was attempted** (intent in one or two sentences), and optionally **what to try next**. This file is append-only; do not delete past entries. Commit this log update with the **next** green commit if it is part of the same work, or as its own tiny commit if the user prefers a clean separation.

Do **not** accumulate uncommitted red edits “to fix in the next message.” Red means revert first, then log.

## Workflow loop

```
edit (tiny) → run tests → pass? → git add + git commit
                      ↘ fail? → revert working tree → append tcr-failure-log.md → try again
```

## Git behavior

- **Commit**: Include only files that belong to the green step; keep commits small and atomic.
- **Revert on failure**: Prefer discarding uncommitted changes to return to last commit. Do not commit broken states. If a commit was already made by mistake while red, use the project’s agreed recovery (e.g. `git reset --soft HEAD~1` only when the user expects rewriting history—default to revert/discard **before** committing).
- **Branch**: Same rules on feature branches: never push a sequence that leaves mainline patterns broken if your workflow requires linear green history.

## Detecting the test command

Resolve how to test **from the repo**, in order:

1. `package.json` → `scripts.test` (or `npm test` / `pnpm test` / `yarn test`).
2. `Makefile` → `make test` or documented target.
3. `pytest`, `cargo test`, `go test ./...`, Maven/Gradle, etc., from project conventions.
4. If unclear, check `README`, CI config (e.g. `.github/workflows`), or ask once.

Use the same command the project already uses; do not invent a second test runner without reason.

**Full suite:** The goal is “everything this project’s standard test command runs.” If the repo documents a different “full” command (e.g. `make test-all` vs `make test`), use the one that matches CI or README for complete coverage.

## Agent conduct

- **Proactive**: After substantive edits, run the **full** test suite before suggesting the task is done.
- **Honest**: If tests cannot be run in the environment, say so and either request permission to run them or describe exact commands for the user—do not claim TCR compliance without a green run.
- **Scope**: TCR applies to coding tasks that touch behavior covered by tests; pure docs-only changes may skip tests only when they cannot affect execution.

## Optional: user escape hatch

If the user explicitly says to **skip TCR** or **bend** the workflow for a one-off (e.g. WIP checkpoint), follow their instruction for that turn only, then return to TCR unless they say otherwise.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [xpepper](https://github.com/xpepper)
- **Source:** [xpepper/tcr-skill](https://github.com/xpepper/tcr-skill)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-xpepper-tcr-skill-tcr-kentbeck
- Seller: https://agentstack.voostack.com/s/xpepper
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.