Back to skill

Security audit

GitHub Actions Workflow Hardening Audit

Security checks for vulnerabilities and agentic risk

Overview

This skill is a transparent local GitHub Actions workflow audit tool, with no evidence of hidden network access, persistence, credential use, or destructive behavior.

Reasonable to install for lightweight local workflow checks, but do not treat it as a complete GitHub Actions security gate. Its parser can miss some valid YAML forms, so confirm important CI policy decisions with a proper YAML-aware workflow linter or additional review.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/workflow-hardening-audit.sh:134
Finding
Valid GitHub Actions YAML Syntax Can Bypass Workflow Hardening Checks<![CDATA[ ## Vulnerability Details **File Location**: `scripts/workflow-hardening-audit.sh`, lines 134–135, 167–169, and 199–205 **Vulnerability Type**: Incomplete YAML parsing and security-check bypass **Risk Level**: Medium ### Vulnerable Code ```python job_key_re = re.compile(r"^[A-Za-z0-9_.-]+:\s*(#.*)?$") uses_re = re.compile(r"^\s*(?:-\s*)?uses:\s*([^\s#]+)") ``` ```python if re.match(r'^on\s*:\s*$', line): in_on = True on_indent = indent continue ``` ```python if indent == jobs_indent + 2 and job_key_re.match(stripped): current_job = stripped.split(':', 1)[0] total_jobs += 1 job_timeout[current_job] = False job_permissions[current_job] = False job_concurrency[current_job] = False continue ``` ### Technical Analysis The audit implementation treats GitHub Actions workflow files as line-oriented text and attempts to recognize YAML structures with regular expressions and fixed indentation assumptions. This does not cover all valid YAML representations. The following valid forms can evade or alter the intended checks: 1. **Quoted action references** A step such as: ```yaml - uses: "actions/checkout@main" ``` is matched with the closing quote included in the captured value. The extracted ref becomes `main"` rather than `main`, so `classify_ref()` does not identify it as a branch-like floating reference. 2. **Flow-style or inline event declarations** These valid declarations are not recognized by the parser: ```yaml on: [pull_request_target] ``` ```yaml on: pull_request_target ``` The parser only enters event-processing mode when a line exactly matches `on:`. Consequently, event filters may skip applicable workflows, and the additional risk score for `pull_request_target` is not applied. 3. **Valid non-two-space job indentation** Job discovery requires a job key to be indented exactly two spaces beyond `jobs:`: ```python indent == jobs_indent + 2 ``` YA ...[truncated 2376 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Replace the regular-expression parser with a maintained YAML parser and inspect the resulting object structure. 2. Account for YAML 1.1 parsers that may interpret the unquoted `on` key as a Boolean value, or select/configure a parser that preserves GitHub Actions semantics. 3. Structurally enumerate every key beneath `jobs` rather than relying on a fixed indentation width. 4. Read `uses` values as decoded YAML scalar values so quoted and unquoted references are classified consistently. 5. Support all valid event declaration forms, including scalar, flow-sequence, and mapping forms: ```yaml on: push on: [push, pull_request_target] on: pull_request_target: ``` 6. Fail closed when a workflow cannot be parsed or has an unexpected structure. A parse failure should not silently produce a lower score. 7. Add regression tests covering: - Single-quoted and double-quoted `uses` values. - Flow-style and scalar `on` declarations. - Two-space, four-space, and other valid indentation widths. - Quoted YAML keys. - Empty and malformed workflows. - Reusable workflows and local action references. - `pull_request_target` combined with mutable action refs. 8. Where practical, require immutable full commit SHA references for third-party actions instead of only flagging a limited set of branch and major-tag names. ]]>
Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

Lp3

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding
Without declared permissions the skill's intent is opaque and cannot be validated.

Static analysis

No suspicious patterns detected.