Skip to content

testforge

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

testforge

README.md

TestForge: Software Verification Operator

Turn a software change, repository, failing test, or feature requirement into risk-ranked, repository-compatible verification evidence—and a release conclusion that says what is still unsafe to ship.

TestForge is a portable Augment with one self-contained verification operator, one self-contained independent reviewer, progressive testing doctrine, operational artifacts, deterministic Python tools, TypeScript/Python stack guidance, three situated examples, behavioral evaluations, and a fileless fallback. Markdown is the canonical human record; JSON is the canonical machine record.

Start here

  1. Read docs/QUICK-START.md.
  2. Install skills/software-verification and skills/verification-reviewer as complete skill folders, or upload the matching one-skill archives from the repository's claude-ai/ directory.
  3. Invoke $software-verification with whatever you have: a diff, repository, defect, test failure, requirement, or release candidate.
  4. Let TestForge inspect before it questions you. It asks only for decision-critical information it cannot recover.
  5. Run the independent reviewer before accepting a release assessment.

Principal entry points

  • skills/software-verification/SKILL.md — scope through evidence-backed release assessment.
  • skills/verification-reviewer/SKILL.md — independent challenge of the evidence chain.
  • assets/templates/verification-manifest.json — canonical working record.
  • scripts/ — repository inspection, stack detection, diff summary, validation, normalization, smell scanning, and report assembly.
  • examples/ — complete TypeScript API, Python regression, and parser-edge demonstrations.
  • evals/ — transfer-oriented cases for risk, oracle, evidence, triage, and safety behavior.
  • fallback/master-prompt.md — copy-paste operation when skill loading or repository tools are unavailable.

Honest capability boundary

TestForge uses the files and tools a host actually exposes. It can create compatible tests and run repository-local checks when execution is available and authorized; it cannot grant file access, install dependencies, create environments, certify compliance, prove defect-freedom, or assume release authority. Active security testing, destructive operations, production access, dependency installation, production-code modification, CI changes, and external publication remain human-authorized actions.

See docs/CAPABILITY-MATRIX.md, docs/LIMITATIONS.md, and SECURITY.md for the exercised boundary. See docs/HOST-COMPATIBILITY.md for the exact Codex, Claude.ai, and Claude Code installation units.