PTSD

PRD → Tests → Implementation. Three pipeline profiles. Any AI tool. Any language.

CLI tool that enforces structured AI development.
Pipeline profiles: full, standard, lite. Choose what fits each feature.
Go. Single binary. Zero dependencies. Claude Code, OpenCode, Cursor.

Website · GitHub · Benchmarks

---

Install

Requires Go 1.25+

go install github.com/veschin/ptsd/v2/cmd/ptsd@latest

If @latest returns a stale version (Go proxy caches ~30 min), pin the tag:

go install github.com/veschin/ptsd/v2/cmd/ptsd@v2.0.0

Verify: ptsd version

Shell completion (optional):

# bash -- add to ~/.bashrc
eval "$(ptsd completion bash)"

# fish -- add to ~/.config/fish/config.fish
ptsd completion fish | source

Uninstall: rm $(go env GOPATH)/bin/ptsd. Per project: rm -rf .ptsd/ .claude/ .git/hooks/pre-commit .git/hooks/commit-msg

Quick Start

mkdir my-project && cd my-project
git init
ptsd init --tool claude    # or: opencode, generic
ptsd feature add auth "Auth" --pipeline full
ptsd feature status auth in-progress
ptsd context --agent       # see what to do next

After ptsd init, start a Claude Code session. Hooks fire automatically -- the LLM sees what to do, gets blocked if it tries to skip, and advances stages as it creates artifacts.

Pipeline Profiles

Every feature has a pipeline profile. Default: standard.

Profile	Stages	Use For
full	PRD -> Seed -> BDD -> Tests -> Impl	Complex, data-heavy features
standard	PRD -> BDD -> Tests -> Impl	Default. Most features
lite	PRD -> Tests -> Impl	Simple utilities, config

Tool Integration

Tool	Enforcement
Claude Code	Full: gate, track, context via .claude/hooks/
OpenCode	Full: via .opencode/plugins/ptsd.ts
Generic	Advisory + git hooks

Polyglot

Test detection: Go, TypeScript, JavaScript, Python, Ruby, Rust, Java, C#.

Test runner auto-detected from project files (package.json, go.mod, pyproject.toml, Cargo.toml, etc).

Commands

ptsd init [--tool claude|opencode|generic]
ptsd feature add <id> <title> [--pipeline full|standard|lite]
ptsd feature list
ptsd feature status <id> <status>
ptsd context --agent
ptsd status
ptsd validate
ptsd review <feature> <stage> <score>
ptsd test map --feature <id> <test-file>
ptsd adopt [--dry-run]
ptsd migrate

Benchmark

8 runs, 2 models (Sonnet, GLM-5 via GoLeM), blind evaluation on identical spec.

Model	Condition	Score	Tests	Commits
Sonnet	bare	25.0/27	22	0-1
Sonnet	PTSD	24.5/27	26	4-8
GLM-5	bare	20.5/27	9	1
GLM-5	PTSD	20.0/27	24	8-13

PTSD doesn't make the first implementation better -- it makes the project developable. After a bare run you have code. After a PTSD run you have a feature registry, BDD specs, test mappings, review history, and stage-separated commits.

Full methodology: FEEDBACK.md | Website

Exit Codes

Code	Meaning
0	Success
1	Validation failure
2	Bad arguments
3	Config error
4	I/O error
5	Test runner failure

License

MIT