tbdflow
Manage Trunk-Based Development workflows using the tbdflow CLI. Use this skill to create short-lived branches, make standardised commits, sync with trunk, merge completed work, and generate changelogs.
About this skill
This AI agent skill enables the management of a Trunk-Based Development (TBD) workflow through the `tbdflow` command-line interface. It's designed to streamline common Git operations by enforcing best practices such as short-lived branches and standardized commit messages, reducing the cognitive load on developers and ensuring a clean, fast path back to the `main` branch. The skill acts as the sole interface for these specific Git workflow actions. Developers can leverage this skill when starting new tasks, committing changes, synchronizing their local environment with the central `main` branch, merging completed features, or generating summaries of recent changes. By abstracting away complex Git commands, it helps maintain consistency and efficiency in teams practicing TBD, preventing common errors and promoting a smooth development cycle.
Best use case
The primary use case for this skill is to automate and standardize Git workflows within a Trunk-Based Development environment. It benefits development teams aiming for continuous integration, rapid iteration, and high code quality by ensuring consistent branching, committing, and merging practices, thus reducing manual Git decision-making and potential human error.
Manage Trunk-Based Development workflows using the tbdflow CLI. Use this skill to create short-lived branches, make standardised commits, sync with trunk, merge completed work, and generate changelogs.
Users can expect a streamlined and standardized Git workflow, with the agent performing common TBD operations efficiently and compliantly.
Practical example
Example input
Hey AI, I'm starting work on ticket #123, can you set me up? Then, commit my staged changes with the message 'feat: Implement user authentication.'
Example output
Branch 'feature/123-user-authentication' created and checked out. Your changes have been successfully committed. What's next?
When to use this skill
- Start work on a new task, ticket, or feature.
- Commit staged changes with standardized messages.
- Sync your local branch with the main trunk or check repository status.
- Merge completed work back to the `main` branch.
When not to use this skill
- Create or manage long-lived feature branches.
- Perform manual Git commands directly, outside of `tbdflow`.
- Rewrite Git commit history or perform interactive rebases.
- Merge work without explicit user confirmation or intent.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/tbdflow/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How tbdflow Compares
| Feature / Agent | tbdflow | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | medium | N/A |
Frequently Asked Questions
What does this skill do?
Manage Trunk-Based Development workflows using the tbdflow CLI. Use this skill to create short-lived branches, make standardised commits, sync with trunk, merge completed work, and generate changelogs.
How difficult is it to install?
The installation complexity is rated as medium. You can find the installation instructions above.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
Related Guides
Top AI Agents for Productivity
See the top AI agent skills for productivity, workflow automation, operational systems, documentation, and everyday task execution.
AI Agents for Coding
Browse AI agent skills for coding, debugging, testing, refactoring, code review, and developer workflows across Claude, Cursor, and Codex.
Cursor vs Codex for AI Workflows
Compare Cursor and Codex for AI coding workflows, repository assistance, debugging, refactoring, and reusable developer skills.
SKILL.md Source
# tbdflow Skill
## Overview
This skill enables an AI agent to manage a **Trunk-Based Development (TBD)** workflow using the `tbdflow` CLI (v0.18.2).
The skill exists to:
- Enforce short-lived branches
- Standardise commits
- Reduce Git decision-making
- Maintain a fast, safe path back to trunk (`main`)
`tbdflow` is the **only interface** the agent should use for Git workflow actions covered by this skill.
---
## When to Use
Use this skill when the user wants to:
- Start work on a task, ticket, or feature
- Commit staged changes
- Sync with trunk or check repository status
- Merge completed work back to `main`
- See what has changed since the last release
- Log a design decision, failed attempt, or reasoning pivot
Typical trigger phrases include:
- "Start working on…"
- "Commit this"
- "Merge my work"
- "Sync me up"
- "What's new?"
- "Note that…" / "Log this decision"
---
## When *Not* to Use
Do **not** use this skill to:
- Create or manage long-lived branches
- Perform manual Git commands outside `tbdflow`
- Rewrite commit history
- Perform interactive rebases
- Merge without explicit user intent
If an action cannot be performed via `tbdflow`, explain the limitation instead of falling back to raw Git commands.
---
## Prerequisites
Before invoking any `tbdflow` command, the agent **must** verify that the CLI is installed and accessible.
### Installation Check
Run the following to confirm availability:
```bash
command -v tbdflow && tbdflow --version
```
If `tbdflow` is **not found**, the agent must attempt to install it automatically using one of the strategies below.
### Installation Strategies (in order of preference)
**1. Cargo Install (requires Rust toolchain)**
If `cargo` is available on the system:
```bash
cargo install tbdflow
```
This downloads the latest release from [crates.io](https://crates.io/crates/tbdflow) and compiles it locally.
**2. GitHub Releases (pre-built binary)**
If `cargo` is not available but `curl` is:
```bash
curl -fsSL https://github.com/cladam/tbdflow/releases/latest/download/tbdflow-$(uname -m)-$(uname -s | tr '[:upper:]' '[:lower:]') -o /usr/local/bin/tbdflow
chmod +x /usr/local/bin/tbdflow
```
Adjust the binary path if `/usr/local/bin` is not writable (e.g. use `~/.local/bin`).
**3. Manual Prompt**
If neither strategy is viable, inform the user:
> `tbdflow` is not installed. Please install it using one of:
> - `cargo install tbdflow`
> - Download a binary from https://github.com/cladam/tbdflow/releases
>
> See the README for details.
### Post-Install Verification
After installation, always confirm:
```bash
tbdflow --version
```
If the version is outdated, suggest:
```bash
tbdflow update
```
---
## Instructions
Follow the instructions below exactly. Each capability defines intent, constraints, and decision rules.
---
### 1. Standardised Committing
**Intent**
Create a structured, conventional commit on trunk or a short-lived branch.
**Staging Behaviour**
* `tbdflow` automatically stages the relevant changes when committing
* The agent must **not** run `git add` or any raw Git staging commands
* No explicit staging step is required from the user or the agent
**Preconditions**
* The working tree contains changes intended for commit
* No unresolved merge conflicts are present
**Command**
```bash
tbdflow commit -t <type> [-s <scope>] -m "<message>" [--issue <issue>] [-b]
```
**Decision Rules**
* Allowed commit types:
* `feat`, `fix`, `chore`, `docs`, `refactor`, `test`, `build`, `ci`, `perf`, `revert`, `style`
* Never invent new commit types
* If no type is specified:
* Default to `chore` unless behaviour changes
* DoD Checklist: If a `.dod.yml` file exists in the project root and `--no-verify` is not passed, an interactive
checklist will appear. Unchecked items will result in a `TODO:` footer being appended to the commit message.
* Use `-b` / `--breaking` if the change introduces breaking behaviour
* Use `--issue` when the user references a ticket ID (JIRA, GitHub, etc.)
**Use This When**
* The user says “commit”, “save this”, or “check this in”
* The user describes completed work ready to be recorded
---
### 2. Creating Short-Lived Branches
**Intent**
Start a new unit of work in a short-lived branch that will be merged back to trunk quickly.
**Preconditions**
* Working tree is clean or safely stashed
* The task is not exploratory or long-running
**Command**
```bash
tbdflow branch -t <type> -n <name> [--issue <issue>] [-f <from_commit>]
```
**Decision Rules**
* Branch naming follows:
* `<type>/<name>` or
* `<type>/<issue>-<name>`
* If the user provides a task description:
* Slugify it for the `-n` parameter
* Use `--issue` when a ticket ID is available
* Use `-f` only if the user explicitly asks to branch from a non-HEAD commit
**Use This When**
* The user says “start working on…”
* A task requires isolation before merging to `main`
---
### 3. Workflow Completion & Integration
**Intent**
Safely merge completed work back into trunk and clean up the branch.
**Preconditions**
* All intended commits are complete
* The branch is ready to be merged
**Command**
```bash
tbdflow complete -t <type> -n <name>
```
**Decision Rules**
* Infer `<type>` and `<name>` from:
1. The current Git branch name
2. Fallback: the most recent `tbdflow branch` invocation
* The merge is performed using `--no-ff`
* Both local and remote branch copies are deleted after merge
**Use This When**
* The user says “I’m done”, “merge my work”, or “ship this”
---
### 4. Syncing & Status
**Intent**
Keep the local workspace aligned with trunk and provide situational awareness.
**Commands**
```bash
tbdflow sync
tbdflow status
```
**Decision Rules**
* Use `sync` to:
* Pull and rebase from remote
* Inspect recent history
* Identify stale branches
* Use `status` to:
* Show context-aware Git status, In monorepos, this excludes sub-project directories when at the root.
* Handle monorepos correctly
**Use This When**
* The user says "sync", "catch me up", or "what's happening"
* Before committing, merging, or starting new work
---
### 5. Radar — Overlap Detection
**Intent**
Proactively detect potential merge conflicts by scanning active remote branches for overlapping work with local changes.
The social coding safety net for TBD.
**Commands**
```bash
tbdflow radar
```
**Decision Rules**
* Use `radar` to:
* Scan all active (unmerged) remote branches
* Compare their diffs against local uncommitted/staged changes
* Show who is working on overlapping files (and optionally overlapping lines)
* Provide actionable social coordination hints
* Radar is also integrated into:
* `tbdflow sync` — shows a one-liner warning if overlap is detected
* `tbdflow commit` — optionally warns or prompts for confirmation
**Detection Levels**
| Level | What it checks | Speed |
|--------|---------------------------------------|--------------|
| `file` | Same files touched (default) | ~5ms/branch |
| `line` | Overlapping line ranges in same files | ~50ms/branch |
**Configuration (`.tbdflow.yml`)**
```yaml
radar:
enabled: true
level: file # file | line
on_sync: true # Show warnings during tbdflow sync
on_commit: warn # off | warn | confirm
ignore_patterns: # Files to exclude from overlap detection
- "*.lock"
- "*-lock.*"
- "CHANGELOG.md"
```
**Use This When**
* The user says "anyone else working on this?", "check for conflicts", or "radar"
* Before pushing to avoid merge hell
* When collaborating closely with teammates on trunk
---
### 6. Undo — The Panic Button
**Intent**
Immediately revert a broken commit on trunk, restoring it to a green state. In TBD, if trunk breaks, you fix it or
revert it — there is no middle ground.
**Command**
```bash
tbdflow undo <sha> [--no-push]
```
**Preconditions**
* The working tree is clean (no uncommitted changes)
* The commit SHA exists and is on the main branch
**Decision Rules**
* `undo` will:
* Sync with remote (fast-forward only) before reverting
* Verify the commit exists and is on trunk
* Create a revert commit using `git revert --no-edit`
* Push the revert to the remote (unless `--no-push`)
* Use `--no-push` when the user wants to inspect the revert locally before pushing
* The reverted changes remain in Git history and can be re-applied later
* This command only works on commits that are on the main branch
**Use This When**
* The user says "revert this", "undo that commit", or "trunk is broken"
* A commit on trunk caused a build failure, test regression, or production incident
* The fastest path to green is reverting rather than fixing forward
---
### Pre-Commit Workflow
**Always run `tbdflow sync` before `tbdflow commit`.**
The `sync` command:
* Pulls and rebases from remote
* Shows current status (wraps `git status`)
* Ensures the workspace is aligned with trunk
This prevents conflicts and ensures commits are based on the latest trunk state.
---
## Validation & Linting Behaviour
`tbdflow` enforces workflow correctness using an internal linter. The agent must understand and respect these rules.
---
### Commit Message Rules
#### Subject Line (`-m` message)
| Rule | Requirement | Example |
|----------------|--------------------------------------------------------------------------------------------------------------|--------------------------------------|
| Max Length | 72 characters | `"add user profile"` ✓ |
| Capitalisation | Must not start with a capital letter | `"add feature"` ✓, `"Add feature"` ✗ |
| Punctuation | Must not end with a period | `"fix bug"` ✓, `"fix bug."` ✗ |
| Type | Must be one of: `feat`, `fix`, `chore`, `docs`, `refactor`, `test`, `build`, `ci`, `perf`, `revert`, `style` | `feat` ✓, `feature` ✗ |
| Scope | Optional, lowercase, no spaces | `-s login` ✓, `-s "user login"` ✗ |
| Message | Required, non-empty, imperative mood | `"add user profile"` ✓, `""` ✗ |
| Breaking | Must use `-b` flag if breaking change | `-b` for breaking ✓ |
#### Commit Body (Optional)
| Rule | Requirement |
|-------------|------------------------------------------------|
| Line Length | Each line must not exceed 80 characters |
| Separation | Must be separated from subject by a blank line |
#### Issue Key (`--issue`)
| Rule | Requirement | Example |
|--------|-------------------------------------|----------------------------|
| Format | Uppercase project key, dash, number | `PROJ-123` ✓, `proj-123` ✗ |
---
### Branch Name Rules
| Rule | Requirement | Example |
|-------|----------------------------------------|--------------------------------------------|
| Type | Must match commit types | `feat/` ✓, `feature/` ✗ |
| Name | Lowercase, hyphen-separated, no spaces | `add-user-profile` ✓, `Add User Profile` ✗ |
| Issue | Optional, prefixed to name | `feat/API-456-add-user` ✓ |
---
### Error Handling
If `tbdflow` rejects input:
1. Read the error message carefully
2. Correct the input based on the rules above
3. Retry with valid parameters
4. Do **not** fall back to raw Git commands
The agent should prefer generating valid inputs over relying on linter errors.
---
### 7. Breadcrumbs — Low-Friction Intent Logging
**Intent**
Capture architectural decisions, failed attempts, or logic pivots during the development process before the final
commit.
Breadcrumbs provide a lightweight way to document the *why* behind code changes — the reasoning that would otherwise be
lost between keystrokes.
**Command**
```bash
tbdflow + "<breadcrumb_message>"
```
**Decision Rules**
* Use this whenever you pivot your approach (e.g. switching from a Factory to a Trait)
* Use this to document a failed attempt that explains why the final code looks the way it does
* Use this to explain complex logic, patterns, or non-obvious trade-offs
* Do **not** use this for trivial changes (typos, formatting)
* These notes are automatically appended to the next `tbdflow commit` body
* Multiple breadcrumbs accumulate and are included in order
**Storage & Safety**
Breadcrumbs are stored locally in `.tbdflow-intent.json` at the repository root. This file is **never committed** — it
is consumed and deleted after the next `tbdflow commit`.
`tbdflow` automatically ensures `.tbdflow-intent.json` is listed in `.gitignore` the first time a breadcrumb is saved.
This prevents the `git add .` staging pattern from accidentally committing raw intent data.
If `.gitignore` does not exist, it will be created with the entry. If it already contains the entry, no changes are
made.
**Decision Rules: The Intent Log**
Breadcrumbs are not optional decoration — they are the agent's **audit trail**.
* **Log your struggle**: if you try an approach and it fails or feels wrong, log it with `tbdflow +`.
This gives the human auditor context on what you *didn't* do and why.
* **Vibe-coding transparency**: do not just push a final "perfect" diff.
Use breadcrumbs to show the evolution of the logic — the dead ends, the trade-offs, the pivots.
* **Minimum expectation**: before committing a complex refactor or non-trivial feature, there should be
at least 1–2 breadcrumbs explaining the major structural decisions.
* **Observability over perfection**: a clean diff with zero context is harder to review than a
slightly messier one with a clear reasoning trail.
> *"Velocity is nothing without observability. The trunk is live, the struggle is logged, the audit is parallel."*
**Agent Behaviour**
* **Think out loud**: use `tbdflow +` to document your chain of thought during development
* Prefer clarification over assumptions for destructive actions
* Default to the safest operation that preserves trunk integrity
* Treat breadcrumbs as a first-class part of the workflow, not an afterthought
**Examples**
| Scenario | Command |
|----------------------------------|--------------------------------------------------------------------------------------------------|
| Switching architectural pattern | `tbdflow + "switched from Factory to Trait: Factory felt over-engineered for this scope"` |
| Documenting a rejected edge-case | `tbdflow + "decided against async here: the overhead outweighs the benefits for this sync task"` |
| Explaining complex regex/logic | `tbdflow + "using lookahead in regex to handle nested brackets without recursion"` |
**Use This When**
* The agent or user makes a non-trivial design decision mid-flight
* A failed approach informs the final implementation
* The commit message alone would not explain the reasoning
---
### 8. Changelog Generation
**Intent**
Summarise changes using structured commit history.
**Command**
```bash
tbdflow changelog [--unreleased] [--from <ref>]
```
**Decision Rules**
* Use `--unreleased` when the user asks "What's new?"
* Use `--from <ref>` when comparing against a specific tag or version
**Use This When**
* The user asks for release notes
* The user wants a summary of recent changes
---
## Output Format
* Commands should be executed directly
* Explanations should be concise and factual
* Avoid narrating Git internals unless asked
* Prefer showing the command being run before or alongside results
---
## Examples
| User Input | Action |
|-----------------------------------------------|-------------------------------------------------------------------------------------------|
| "Commit this as a bug fix for login." | `tbdflow commit -t fix -s login -m "resolve timeout issue"` |
| "Start working on API-456: Add user profile." | `tbdflow branch -t feat -n add-user-profile --issue API-456` |
| "Merge my current work back to main." | `tbdflow complete -t <current_type> -n <current_name>` |
| "Sync me up." | `tbdflow sync` |
| "Anyone else working on this file?" | `tbdflow radar` |
| "Revert commit abc1234, it broke the build." | `tbdflow undo abc1234` |
| "I switched from Factory to Trait." | `tbdflow + "switched from Factory to Trait: Factory felt over-engineered for this scope"` |
| "What changed since the last version?" | `tbdflow changelog --unreleased` |
---
## Notes
* Treat `main` (trunk) as sacred
* Prefer safety and clarity over cleverness
* Ask for clarification only when an action could be destructiveRelated Skills
linux-shell-scripting
Provide production-ready shell script templates for common Linux system administration tasks including backups, monitoring, user management, log analysis, and automation. These scripts serve as building blocks for security operations and penetration testing environments.
iterate-pr
Iterate on a PR until CI passes. Use when you need to fix CI failures, address review feedback, or continuously push fixes until all checks are green. Automates the feedback-fix-push-wait cycle.
istio-traffic-management
Comprehensive guide to Istio traffic management for production service mesh deployments.
incident-runbook-templates
Production-ready templates for incident response runbooks covering detection, triage, mitigation, resolution, and communication.
incident-response-smart-fix
[Extended thinking: This workflow implements a sophisticated debugging and resolution pipeline that leverages AI-assisted debugging tools and observability platforms to systematically diagnose and res
incident-responder
Expert SRE incident responder specializing in rapid problem resolution, modern observability, and comprehensive incident management.
expo-cicd-workflows
Helps understand and write EAS workflow YAML files for Expo projects. Use this skill when the user asks about CI/CD or workflows in an Expo or EAS context, mentions .eas/workflows/, or wants help with EAS build pipelines or deployment automation.
error-diagnostics-error-trace
You are an error tracking and observability expert specializing in implementing comprehensive error monitoring solutions. Set up error tracking systems, configure alerts, implement structured logging,
error-debugging-error-trace
You are an error tracking and observability expert specializing in implementing comprehensive error monitoring solutions. Set up error tracking systems, configure alerts, implement structured logging, and ensure teams can quickly identify and resolve production issues.
error-debugging-error-analysis
You are an expert error analysis specialist with deep expertise in debugging distributed systems, analyzing production incidents, and implementing comprehensive observability solutions.
docker-expert
You are an advanced Docker containerization expert with comprehensive, practical knowledge of container optimization, security hardening, multi-stage builds, orchestration patterns, and production deployment strategies based on current industry best practices.
devops-troubleshooter
Expert DevOps troubleshooter specializing in rapid incident response, advanced debugging, and modern observability.