本文へ移動
cccskills
無料GitHub で公開

bdb-security-audit

Comprehensive security auditing, diff analysis, defensive checklists, and vulnerability testing.

インストール方法を見る

含まれるファイル(4)

  • SKILL.md5.1 KB
  • references/audit-checklists.md13.8 KB
  • references/skylos-guide.md5.9 KB
  • references/xalgorix-bridge.md5.0 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

BDB Security Audit

Defensive security auditing for BDB projects. Use before commits, during PR review, inside the Shipping quality gate, or whenever a codebase needs a structured security pass. This skill is defensive only: it verifies, hardens, and reports. It never produces attack tooling, exploit payloads, or offensive reconnaissance.

The Three Pillars

PillarWhat it isWhen it runs
SkylosDiff and AST-level security analyzer for code changes. Catches dangerous call patterns, hardcoded credentials, and dead code that carries secrets — on staged files and PR diffs, before they land.Pre-commit, PR review, Shipping gate
Claude-Red ChecklistsStructured defensive audit checklists derived from red-team knowledge, inverted into verification questions. Covers OWASP Top 10, AuthN/AuthZ, injection classes, SSRF, LLM/agent security, and secret leakage.Deep audits, new-feature review, release prep
Xalgorix BridgeDecoupled connection to the xalgorix automated security suite. Discovered at runtime via $PATH or XALGORIX_URL — never hardcoded. Runs broader automated scans and returns findings.Scheduled scans, pre-release sweeps, CI

The pillars are independent. A machine may have none, one, or all three; every workflow below degrades gracefully and must report "could not check" rather than "checked, fine" when a tool is absent.

When to Use

  • "Audit this before I push" — run Skylos on the diff, then the relevant checklists.
  • "Security review this PR" — skylos diff against the base branch, triage findings.
  • Shipping gate step — Skylos must pass (or every finding must be dispositioned) before a release.
  • "Is this endpoint/agent/tool safe?" — pull the matching section from references/audit-checklists.md.
  • Periodic automated sweep — Xalgorix Bridge, if the suite is reachable.

Workflow

1. Diff pass (always first, always cheap)

# Staged / uncommitted changes
skylos diff

# Whole tree, if no baseline exists
skylos scan .

Verify the binary exists first — command -v skylos — and check skylos --help for the flags your installed version supports. If skylos is absent, record "could not check" and continue; do not silently treat the step as passed.

2. Checklist pass (targeted, human-driven)

Open references/audit-checklists.md and work only the sections the change touches: touched auth code → Authentication & Session Management; new endpoint → Input Validation + Infrastructure; new agent/tool surface → LLM & Agent Security. Every item is phrased as a verification question. Mark each pass / fail / not applicable — an unmarked item is not a pass.

3. Automated sweep (optional, decoupled)

# Discover the suite — never assume, never hardcode
command -v xalgorix || echo "not on PATH"
echo "${XALGORIX_URL:-not configured}"

If neither is present, skip and report. If present, run the scan per references/xalgorix-bridge.md and forward findings to Reviewer/Shipping.

4. Report

Consolidate into one summary for Reviewer or Shipping:

  • Per finding: file, line, severity, checklist or scanner origin, disposition (fix now / ticket / accepted risk).
  • Per check: pass, fail, or could not check — never leave a check unreported.
  • A gate verdict only from evidence: a passing report with two "could not check" rows is not a green gate.

Commands Quick Reference

command -v skylos            # pillar 1 present?
skylos diff                  # security scan of staged/changed code
skylos scan .                # full-tree scan
skylos --help                # confirm flags for your installed version

command -v xalgorix          # pillar 3 on PATH?
echo "$XALGORIX_URL"         # pillar 3 via remote endpoint?

Guardrails

  1. Defensive posture only. Findings describe what is wrong and how to fix it. Never write or store exploit payloads, working attack strings, or recon steps — not even in reports, tickets, or memory.
  2. No hardcoded paths. Tools are discovered via $PATH or environment variables (XALGORIX_URL). Skills, agents, and configs never contain absolute machine paths.
  3. No secret leakage. Audit outputs must redact credentials, tokens, and keys found during scanning. Name the file and line, never the value.
  4. Honest gates. "Could not check" is a distinct, reported state. Silent degradation to "pass" is the defect class this skill exists to prevent.
  5. English only, minimal scope. Reports state findings and fixes; they do not lecture, and they do not expand scope beyond the diff or audit target.

References

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Use when manage multiple local CLI agents via tmux sessions (start/stop/monitor/assign) with cron-friendly scheduling.

日本語の概要は準備中です。原文の説明を表示しています。

hybridlabor-api/aos62026年10月11日 更新

Use when a hybrid memory system that provides persistent, searchable knowledge management for AI agents (Architecture, Patterns, Decisions).

日本語の概要は準備中です。原文の説明を表示しています。

hybridlabor-api/aos62026年10月11日 更新

Reference for how BDB structures autonomous software engineering work — the seven-node dispatcher graph (Architect, TechLead, UI/UX, Engineering, Media/EventTech, Reviewer, Shipping) that /startcycle-graph actually runs. Use when you need the high-level lifecycle framing without inventing your own process.

日本語の概要は準備中です。原文の説明を表示しています。

hybridlabor-api/aos62026年10月11日 更新

Tools are how AI agents interact with the world. A well-designed tool is the difference between an agent that works and one that hallucinates, fails silently, or costs 10x more tokens than necessary. This skill covers tool design from schema to error handling.

日本語の概要は準備中です。原文の説明を表示しています。

hybridlabor-api/aos62026年10月11日 更新

Harness patterns for coding agents — memory, permissions, context engineering, delegation, skills, hooks, bootstrap.

日本語の概要は準備中です。原文の説明を表示しています。

hybridlabor-api/aos62026年10月11日 更新

agenttrail: live map of a multi-agent build in the browser: which plan component is being worked on, by which agent or harness, what is done and what is stuck. Use when a multi-agent pipeline starts (/startcycle, /startcycle-graph, /teamwork-preview) or after a plan-canvas approve, when the user asks to see what the agents are doing ("live map", "build board", "who is running now"), or by yourself whenever a multi-agent run is under way — no user prompt needed.

日本語の概要は準備中です。原文の説明を表示しています。

hybridlabor-api/aos62026年10月11日 更新

hybridlabor-api のスキルをすべて見る

このスキルの問題を報告する