Skip to content

README.md

Latest commit

 

History

History
43 lines (34 loc) · 1.81 KB

README.md

File metadata and controls

43 lines (34 loc) · 1.81 KB
Raw

Agent scripts

Reusable prompts, role contracts, evaluation sets, and operating playbooks for software agents. This directory is for instructions that an agent reads. The repository's root scripts/ directory remains reserved for executable build and maintenance tooling.

Each script is a self-contained folder:

agent-scripts/<script-name>/
  README.md          # purpose, installation, and operating notes
  SYSTEM_PROMPT.md   # canonical vendor-neutral instruction
  roles/             # optional role-specific overlays
  evals/             # behavioral regression and capability cases

Available scripts

Add a script

  1. Create one kebab-case folder with a canonical SYSTEM_PROMPT.md.
  2. Write a README that states the trigger, inputs, installation options, permissions, and verification method.
  3. Add role overlays only when responsibilities genuinely differ.
  4. Add regression cases before tuning the prompt. Include cases where the agent should act and cases where it must not act.
  5. Keep model IDs and vendor-specific configuration outside the evergreen prompt. Document optional adapters separately.
  6. Add an optional reader-facing field note under portfolio/content/writing/ and rebuild the portfolio.

When several harnesses can consume the same prompt, include a short task packet that users can give their existing agent to install it. Tell that agent to use native global instructions, preserve existing guidance, keep optional roles out of always-on context, and verify the result.

Prompt text is policy, not enforcement. Sandboxing, network restrictions, protected paths, approval gates, and audit logging belong in the agent harness.