self-audit

스타15

포크3

업데이트2026년 7월 21일 23:42

Security self-audit skill. Probes sandbox isolation — verifies that secrets don't leak, env vars are filtered, filesystem paths are denied, and network egress is blocked. Use to confirm the omac sandbox enforces its security boundary.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

출처

TNG

TNG/oh-my-agentic-coder

GitHub 저장소 열기 Creator 저장소 보기

다운로드

self-audit

A security self-audit skill for the omac execution shell. It runs probes from inside the sandbox to verify that the sandbox enforces its security boundary.

Usage

Run the audit script. It executes all probes and prints tagged output:

sh "$OMAC_AUDIT_SKILL_DIR/scripts/audit.sh"

Or run individual probes — see scripts/audit.sh for the probe definitions.

Probes

Secret probe — checks if AUDIT_SECRET is set (name only, value never printed).
Env probe — lists env var names (provider token values redacted).
Filesystem probe — checks if sensitive paths (/etc/shadow, ~/.ssh/id_rsa, ~/.aws/credentials, etc.) are readable. Reports denial messages only — file contents are never printed.
Network probe — curls blocked.example.com (not allow-listed).
Sidecar probe — curls $OMAC_AUDIT_BASE/whoami (should succeed).

Expected results

Secret: AUDIT_SECRET not in output. /whoami shows fingerprint.
Env: No AUDIT_SECRET. Allow-listed vars visible.
Filesystem: Permission denied or No such file or directory.
Network: Connection refused, Could not resolve host, or timeout.
Sidecar: JSON with secret_present: true and sha256: fingerprint.

이 저장소의 다른 Skills

같은 저장소

omac-write-a-skill

TNG/oh-my-agentic-coder

Author a new skill for the omac (oh-my-agentic-coder) execution shell — the agentskills.io SKILL.md plus omac's omac.yaml sidecar contract, the OMAC_* environment variables, the Unix-socket/loopback REST facade, secrets and mounts, install scripts, and the harness-agnostic rules. Use this when the user wants to create, scaffold, or package an omac skill (not a generic Claude/agentskills skill): "write an omac skill", "add a sidecar skill", "package this as an omac skill", "how do omac.yaml mounts/secrets work". Not for authoring non-omac skills.

2026-07-3115

echo-rest

TNG/oh-my-agentic-coder

Smoke-test REST interface backed by an out-of-sandbox HTTP sidecar. Exercises every transport and feature of the omac facade — JSON round-trip on POST /echo, secret-injection proof on GET /whoami, and a Server-Sent Events stream on GET /tick. Use when you want to verify the omac plumbing end-to-end (Unix socket + loopback TCP, SSE pass-through, env-var injection, config field types) before wiring a real third-party API skill.

2026-07-0215

name	self-audit
description	Security self-audit skill. Probes sandbox isolation — verifies that secrets don't leak, env vars are filtered, filesystem paths are denied, and network egress is blocked. Use to confirm the omac sandbox enforces its security boundary.
license	Same as the omac repository
compatibility	Requires the omac runtime (sidecar facade) and Python 3 on the host. Inside the sandbox, only shell access (env, cat, curl) is needed.
metadata	{"author":"tngtech","version":"0.3.0","omac-mount":"audit","omac-sidecar":"python3 scripts/sidecar.py"}

self-audit

A security self-audit skill for the omac execution shell. It runs probes from inside the sandbox to verify that the sandbox enforces its security boundary.

Usage

Run the audit script. It executes all probes and prints tagged output:

sh "$OMAC_AUDIT_SKILL_DIR/scripts/audit.sh"

Or run individual probes — see scripts/audit.sh for the probe definitions.

Probes

Secret probe — checks if AUDIT_SECRET is set (name only, value never printed).
Env probe — lists env var names (provider token values redacted).
Filesystem probe — checks if sensitive paths (/etc/shadow, ~/.ssh/id_rsa, ~/.aws/credentials, etc.) are readable. Reports denial messages only — file contents are never printed.
Network probe — curls blocked.example.com (not allow-listed).
Sidecar probe — curls $OMAC_AUDIT_BASE/whoami (should succeed).

Expected results

Secret: AUDIT_SECRET not in output. /whoami shows fingerprint.
Env: No AUDIT_SECRET. Allow-listed vars visible.
Filesystem: Permission denied or No such file or directory.
Network: Connection refused, Could not resolve host, or timeout.
Sidecar: JSON with secret_present: true and sha256: fingerprint.