autospec-secaudit

name	autospec-secaudit
description	Use when the user wants to scan generated code for security vulnerabilities, secret/credential leaks, IP/copyright/license violations, PII/data leaks, SQL injection, prompt injection, and backdoors. Runs as `/autospec-secaudit` (manual repo-wide sweep) or auto-fires after each autospec-run batch unless `~/.autospec/no-secaudit.flag` exists. Also powers the blocking Phase 4 per-PR gate.

Self-update mode

Decide this purely from the request text the harness handed you. Do NOT shell out to test the user's free-form request. Read the request, normalize it in your reasoning (collapse whitespace, trim, lowercase), and if the result is exactly update, this skill enters self-update mode and does NOT run the normal pipeline.

1. Detect harness by checking which install path exists for this skill:
   • Claude Code: ~/.claude/skills/autospec-secaudit/SKILL.md
   • OpenCode: ~/.config/opencode/agent/autospec-secaudit.md
   • Codex CLI: ~/.codex/prompts/autospec-secaudit.md
2. Re-install from main by piping the canonical installer:

   curl -fsSL https://raw.githubusercontent.com/berlinguyinca/autospec/main/bootstrap.sh | bash -s -- --skill all --harness all --update
   

   Run this one-liner once; it refreshes all autospec skills across all harnesses.
3. Show the diff between the prior installed file(s) and the freshly fetched copy.

Stop mode

If the normalized request is exactly stop (or matches ^\s*stop(\s+--\w+)*\s*$), do NOT run the pipeline. Write ~/.autospec/stop.flag so any running autospec monitor halts after the current issue, then report that the stop sentinel was written and exit.

Required capabilities & harness adapter

Capability	Claude Code	OpenCode	Codex CLI	Fallback if missing
Subagent model tier	Tier A: opus + ultrathink	Tier A: top-tier task + max reasoning	Tier A: gpt-5.1-codex + reasoning_effort=high	Fall back UP on unavailability
Subagent dispatch policy	per AGENTS.md decision matrix	per AGENTS.md decision matrix	per AGENTS.md decision matrix	inline if subagents unavailable
Deterministic scanners	gitleaks, semgrep, trivy, license-checker via ensure-tool.sh	same	same	LLM-only fallback per missing scanner (loud WARN)

Triage runs on the spec-work tier. Model tier: TIER_A — confirm/triage findings with the top-tier reasoning model for the active harness (see row above).

Harness detection (run once at skill start, before the pipeline)

Detect your harness by checking available tools before scanning:

1. Claude Code — the Agent tool with a subagent_type parameter is available.

   • TIER_A = opus + ultrathink (model ID: claude-opus-4-7)
   • TIER_B = sonnet (model ID: claude-sonnet-4-6)

2. OpenCode — a task tool with model/tier configuration is available (no subagent_type).

   • TIER_A = top-tier task model + high reasoning
   • TIER_B = smaller-tier task model + medium reasoning

3. Codex CLI — neither Agent nor a configurable task tool is available; apply_patch is the primary edit tool.

   • TIER_A = current top GPT model + reasoning_effort=high
   • TIER_B = gpt-5.1-codex-spark + reasoning_effort=medium

Fallback rule: If TIER_B is not available in your harness (model unknown, quota/capacity failure, authorization failure, or tool call returns an error for that model), silently retry the same dispatch with TIER_A. Preserve the parent context on retry; for Codex native subagents, fork/inherit the current conversation context and use the latest top GPT model instead of moving the work into the main session. Never ask the user.

Security triage runs on TIER_A. Hold TIER_A and TIER_B for the entire skill run.

When to invoke

• Manually: /autospec-secaudit for a repo-wide sweep of the working tree.
• Automatically: auto-fires after each /autospec-run batch unless ~/.autospec/no-secaudit.flag exists.
• As the Phase 4 gate engine (invoked by the implementer, not by a skill call).

Pipeline

1. Ensure scanners — ensure-tool.sh gitleaks semgrep trivy license-checker (best-effort; a failed install degrades that concern to LLM-only with a loud WARN — never silently skipped).
2. Scan — run security-scan.sh --tree (sweep) or --diff <base> (branch). Findings are emitted as gap JSON objects: {gap_id,dimension,severity,file,line,title,body,dedupe_key}.
3. LLM triage (Tier A) — for each finding: confirm real vs false-positive; ADD findings scanners miss (PII in logs, prompt-injection sinks where untrusted input reaches an LLM call, contextual backdoors). Assign final severity (must-fix / nice-to-have).
4. Report — write .autospec/secaudit.md summarizing findings by dimension with file:line and remediation. Loudly list any scanner that fell back to LLM-only.
5. File survivors — must-fix survivors above threshold are filed as auto-implement,security,priority:high issues via the gap-filing path so they re-enter autospec. dedupe_key prevents re-filing across batches.

Enforcement defaults

Concern	dimension	Phase 4 gate	Auto-fix
Secrets / credentials	secrets	block	redact + flag rotation
Vulns / SQLi / cmd-injection / backdoors	vuln	block	yes, re-scan
Prompt injection	injection	advisory (deterministic; LLM triage blocks)	yes, re-scan
PII / data leaks	pii	advisory (deterministic; LLM triage blocks)	yes, re-scan
IP / copyright / license	license	advisory (deterministic; LLM triage blocks)	advisory fix (human confirms)
Dependency CVEs	cve	advisory (block only critical+fixable)	bump if trivial

Finalization

Print the report path and a one-line summary: secaudit: must-fix=<N> advisory=<N> scanners-degraded=<list>.