Strix (STRIX) — skills (internal knowledge packs)

docs / modules / skills

← all docs

Strix (STRIX) — skills (internal knowledge packs)

flowchart LR IN["IN: role (root/child),\nscan_mode, is_whitebox flag"] --> ACT["ACT: resolve a skill list,\nload markdown bodies"] ACT --> VAL["VAL: — none\n(pure file lookup)"] VAL --> OUT["OUT: skill text injected into\nthe rendered system prompt,\nor fetched on demand mid-run"]

What this is

strix/skills/ is a tree of Markdown files — not code — that acts as Strix's methodology library: how to do SQL injection testing, how to drive nuclei, how a root agent should coordinate its children, and so on. It's loaded by strix/agents/prompt.py at prompt-render time and by the load_skill tool mid-run.

This is a different thing from the top-level repo skills/ folder (outside strix/), which holds SKILL.md files for external coding agents (Claude Code, Cursor, Codex) that want to drive Strix from an agentic coding session (penetration-testing-with-strix, ci-security-scanning-with-strix, fix-security-vulnerabilities-with-strix, managed-pentesting-with-strix) — see ../process/00_process-and-runbooks-index.md for that distinction restated for operators.

What flows through it

flowchart TD Prompt["agents/prompt.py\n_resolve_skills()"] -- "always" --> ScanMode["skills/scan_modes/<mode>.md"] Prompt -- "always" --> Tooling["skills/tooling/agent_browser.md\nskills/tooling/python.md"] Prompt -- "if root" --> Coord["skills/coordination/root_agent.md"] Prompt -- "if whitebox" --> SAST["skills/custom/source_aware_sast.md\nskills/coordination/source_aware_whitebox.md"] Prompt -- "caller-requested" --> Any["any skills/<category>/<name>.md"] Runtime["tools/load_skill/tool.py"] -- "mid-run, on demand" --> Any

Technical details

Package: strix/skills/ (pure Markdown, no Python besides __init__.py). Depends on: nothing (leaf content). Depended on by: strix.agents.prompt (always-loaded set, baked into every system prompt), strix.tools.load_skill (on-demand mid-run fetch).

Taxonomy (10 subdirectories)

Directory Count Covers
vulnerabilities/ 25 Per-vulnerability-class methodology — the largest category (SQLi, XSS, IDOR, SSRF, XXE, RCE, JWT auth, race conditions, prompt injection, and more).
tooling/ 11 Per-tool usage guides — includes agent_browser.md and python.md, which are always loaded regardless of scan mode.
technologies/ 5 Auth0, Firebase, Supabase, Active Directory, Grafana/Prometheus.
frameworks/ 4 Django, FastAPI, NestJS, Next.js.
cloud/ 3 AWS, GCP, Kubernetes.
custom/ 3 source_aware_sast.md (whitebox-only), api_spec_testing.md, dependency_cve_scanning.md.
scan_modes/ 3 quick.md, standard.md, deep.md — one is always auto-loaded based on --scan-mode.
protocols/ 2 GraphQL, OAuth.
coordination/ 2 root_agent.md (orchestration guidance, root-only), a peer-coordination doc.
reconnaissance/ 1 Asset discovery.

Related: STRIX — Backends, agents.md, ../design/HLD.md.