wtf

A CLI tool that explains why your command just failed. Wraps any command, and when it exits non-zero, uses an LLM + Stack Overflow to diagnose the error.

How it works

wtf <command> [args...]

1. Runs your command, passing through stdout/stderr in real time
2. If the command succeeds (exit 0), wtf exits silently
3. If it fails, wtf captures stderr and:
   • Splits the output into independent error blocks (via LLM)
   • Searches Stack Overflow for each error
   • Explains each error with root cause, confidence, and actionable next steps

$ wtf cargo build
   Compiling myapp v0.1.0
error[E0382]: borrow of moved value: `x`
  ...

✗ Command failed (exit code 101) — cargo build

━━━ wtf diagnosis ━━━

── Error 1: borrow of moved value ──

Root cause: Variable `x` was moved into a closure and then used again after the move.
Confidence: High — this is a standard Rust ownership error.
Evidence: Multiple SO answers confirm moving into closures transfers ownership.

Next steps:
  • Clone `x` before passing it to the closure
  • Use a reference instead of moving the value

Related Stack Overflow questions:
  → Use of moved value in Rust closure
    https://stackoverflow.com/q/12345678

Install

go install github.com/kht/wtf/cmd/wtf@latest

Or build from source:

make build    # produces ./wtf

Configuration

wtf uses layered configuration: config file → environment variables → CLI flags (highest priority).

Config file

Copy the example and edit:

# Linux/macOS
cp wtf.example.toml ~/.config/wtf/config.toml

# Windows
copy wtf.example.toml %APPDATA%\wtf\config.toml

[llm]
provider = "openai"       # "openai" or "anthropic"
model = "gpt-4.1"
api_key = ""              # or use WTF_LLM_API_KEY env var

# Use different models for splitting vs explaining
[llm.split]
model = "gpt-4.1-mini"   # cheaper model for splitting errors

[llm.explain]
# provider = "anthropic"
# thinking_budget = 10000 # enable extended thinking

[analysis]
timeout_secs = 60
excerpt_chars = 12000
stack_overflow_results = 5

Environment variables

Variable	Description
WTF_LLM_PROVIDER	LLM provider (openai or anthropic)
WTF_LLM_MODEL	Model name
WTF_LLM_API_KEY	API key
WTF_LLM_BASE_URL	Custom API endpoint
WTF_LLM_SPLIT_*	Per-step overrides for the split phase
WTF_LLM_EXPLAIN_*	Per-step overrides for the explain phase
WTF_ANALYSIS_TIMEOUT_SECS	Per-call timeout
WTF_ANALYSIS_EXCERPT_CHARS	Max stderr chars sent to LLM
WTF_ANALYSIS_STACK_OVERFLOW_RESULTS	Max SO results per error

CLI flags

--config          path to config file
--provider        LLM provider
--model           LLM model
--api-key         API key
--base-url        custom API endpoint
--timeout-secs    per-call timeout
--excerpt-chars   max stderr chars sent to LLM
--stack-overflow-results  max SO results per error

License

AGPL-3.0