Skip to content

README.md

Latest commit

 

History

History
63 lines (53 loc) · 3.21 KB

README.md

File metadata and controls

63 lines (53 loc) · 3.21 KB
Raw

f0_sectools documentation

Start from who you are:

flowchart LR
    EV(["Evaluating it?"]) --> SM["explanation/security-model.md<br/>+ the README + demo"]
    OP(["Running it?"]) --> UG["user-guide/ —<br/>getting started → your runtime"]
    SR(["Reviewing it for deployment?"]) --> SM2["explanation/security-model.md<br/>then verify with the tests"]
    CO(["Contributing?"]) --> CT["CONTRIBUTING.md +<br/>explanation/architecture.md"]
    AI(["Here for the small-model thesis?"]) --> TH["explanation/small-model-design.md<br/>+ evals/SCORECARD.md"]
Loading

Understand it (explanation)

  • Architecture — shared core + thin servers, the anatomy of a tool call, and the design decisions behind them.
  • Security model — the trust story: threat model, gated writes, redaction, credential isolation, audit. Read this before deploying; written for the person who approves it.
  • The findings schema — the single output contract, its lifecycle, and the persona renderers.
  • Small-model design — why every tool is flat, bounded, and few; and the eval loop that keeps it measurable.
  • Design history — the dated plan/spec archive (superpowers/) indexed as an ADR log.

Run it (guides)

  • User Guide — setup, per-runtime guides (Hermes, pi, opencode, Claude Code, LM Studio, Open WebUI), skills & personas, workflows, prompting, troubleshooting.
  • Running with local models — serving a tool-calling model with vLLM / llama.cpp.
  • Runtime performance — choosing a serving stack and model: Ollama vs vLLM vs llama.cpp benchmarks.
  • Gated actions how-to — enable, approve (watcher/token/chat), and audit the write actions.
  • FAQ — privacy, hardware, licensing, platform coverage, "can the AI act alone?".
  • Demo — a full tool call end-to-end, offline, in 30 seconds; more artifacts in examples/ (sample findings, annotated transcripts, persona renders).

Look it up (reference)

  • Tool reference — all 51 tools across the 8 servers: descriptions, parameters, enums, defaults, gated-write badges. Generated from the live tool registries; CI fails if it drifts.
  • Skills catalog — all 25 portable skills, by platform. Generated from SKILL.md frontmatter.
  • Glossary — every term, defined once.
  • Eval scorecard — measured model × server callability matrix; agentic eval for multi-step runs.
  • Credentials & permissions per platform: each server's README under servers/ and its .env.<platform>.example.

House rules & history

  • CLAUDE.md — the build guide and critical rules for agents (and humans) working on this repo.
  • proposals/ — accepted documentation plans.