Typesafe Adversary — Development agent for Claude Code
Red-teams TypeSafe integrations before users do.
How to install Typesafe Adversary
Installs to ~/.claude/agents/herakles-dev-typesafe-claude-kit-typesafe-adversary.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/herakles-dev/typesafe-claude-kit/HEAD/agents/typesafe-adversary.md -o ~/.claude/agents/herakles-dev-typesafe-claude-kit-typesafe-adversary.md Restart Claude Code, or start a new session, for it to be picked up.
What Typesafe Adversary does
name: typesafe-adversary description: "Red-teams TypeSafe integrations before users do. Probes the nine documented Jev failure modes, tests prompt injection through state, hunts literal-reading traps and false structural invariants. Produces failing cases with the exact state that triggered them. Required before any integration that reads untrusted text ships." model: opus color: red category: typesafe default_mode: subagent effort: high triggers:
- "break this typesafe integration"
- "r
Alternatives in Development
- Ir Correctness Reviewer — Reviews Slang IR pass changes for correctness, SSA invariants, and type system integrity 5.6k ★
- Cross Linker — Scan recent or specified notes for missing wikilinks 4.6k ★
- Herald — You are a specialized release agent 3.6k ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Bench Reviewer
Reviews eval suite quality — fixture/rubric consistency, scoring accuracy, false positive traps, difficulty ca
Crispener
Refactoring operator that absorbs invariants from business logic into existing schemas — decode/guard-wall del
Typesafe Engineer
Implements TypeSafe integrations: registered QuestionSets, client wiring, answer composition in code, and live
Autorun Shell Reviewer
Reviews shell-script changes under scripts/autorun/ against the documented pitfalls that have caused false-don
PR Review Architecture Maker
Execution-grade PR reviewer scoped to the architecture discipline only — new tradeoffs, module boundaries, rev
Docs Sync
Updates the living specifications (RSM_, TD32_FORMAT_NOTES, DAP_DEBUGGER_ARCHITECTURE, KNOWN_UNKNOWNS, TRAPS,
Related Skills
Axguard Adversary
False Positive Adversary diagnostic (not a vuln report). Usage: /axguard-adversary [path]
Jev Pruner
Claude Code plugin: trim long Bash output with TypeSafe Jev before the model sees it
Canny
Stops AI coding agents from claiming work is done without evidence. Deterministic hooks decide, TypeSafe's Jev