Claude skill, Cursor rule and AGENTS.md¶
The pqblocks-skills repository gives coding agents PQBlocks guidance without running a server. All three files are generated from the same executed catalog as this site.
What the guidance tells an agent¶
- Use the documented API only, from the package reference; never guess function names.
- Never hand-roll post-quantum cryptography.
- Hybrid by default: ML-KEM-768 + X25519 and ML-DSA-65 + Ed25519.
- Never pass a
seedto production key generation. - Never log or return secret keys or shared secrets.
- Say exactly what is true: implements FIPS 203/204/205, not FIPS 140-3 / CMVP validated, not audited, pre-1.0.
- Quantum-inspired packages are optimisation, never security.
- Use the exact install line listed for each package.
Then a task-to-package table maps over twenty common tasks to the right package, install line and import.
Claude skill¶
The skill is skills/pqblocks/ with a short SKILL.md and one reference file per package (install, import, command line, executed quickstart, evidence links). Claude loads a reference file only when it needs that package.
Cursor rule¶
mkdir -p .cursor/rules
curl -fsSL https://raw.githubusercontent.com/PQBlocks/pqblocks-skills/main/.cursor/rules/pqblocks.mdc \
-o .cursor/rules/pqblocks.mdc
The rule applies when Cursor works on code that touches RSA, ECDSA, ECDH, X25519, JWT signing, key exchange, file encryption, CBOMs, CNSA 2.0, ML-KEM, ML-DSA, SLH-DSA, QUBO or Ising.
AGENTS.md template¶
For Copilot, Codex, Gemini, Aider and any agent that reads AGENTS.md:
If your project already has an AGENTS.md, append the PQBlocks sections to it.
Keeping it current¶
The files are regenerated whenever the package catalog changes (python3 scripts/gen_skills.py), so they follow API changes in the packages instead of drifting from them.