Back to skill

Security audit

Changelog Weaver

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly purpose-aligned, but it needs review because untrusted commit text can be fed into AI prompts and generated release Markdown without clear boundaries or escaping.

Before installing, be comfortable with a local script reading repo commit history and writing changelog files you specify. Treat commit messages and AI rewrites as untrusted content, review generated release notes for injected links/headings/instructions, and avoid loading private repository metadata into a tool-enabled chat without an explicit inert-data boundary.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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 (2)

T01 · Skill Instruction Hijacking

Warning
Location
scripts/weaver.py:132
Finding

Untrusted Git Commit Content Is Passed to AI Without Instruction/Data Boundaries

Content
View full analysis
5 else "" cc_match = CC_PATTERN.match(subject) if cc_match: commit_type = cc_match.group("type").lower() scope = cc_match.group("scope") or "" has_breaking = bool(cc_match.group("breaking")) description = cc_match.group("description").strip() else: commit_type = "other" scope = "" has_breaking = False description = subject.strip() breaking_detail = "" if not has_breaking: bm = BREAKING_FOOTER.search(body) if bm: has_breaking = True breaking_detail = bm.group("detail").strip() refs = REF_PATTERN.findall(subject + " " + body) commits.append({ "hash": lines[0], "hash_short": hash_short, "author": author, "email": email, "date": date, "subject": subject, "body": body.strip(), "type": commit_type, "scope": scope, "breaking": has_breaking, "breaking_detail": breaking_detail, "description": description, "references": list(set(refs)), }) ``` The collected fields are copied into the generated JSON: ```python payload["categories"][cat] = { "label": label["category"], "emoj ...[truncated 3686 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/weaver.py:346
Finding

Unescaped Repository Metadata Allows Markdown and Release-Content Injection

Content
View full analysis
1 else "" lines.append(f"- {desc}{refs_str}{count_note}{author_str}") if e.get("breaking") and e.get("breaking_detail"): lines.append(f" - **BREAKING:** {e['breaking_detail']}") lines.append("") ``` The release-note generator performs equivalent interpolation: ```python for e in entries: desc = (ai_rewrites or {}).get(e["hash_short"], e["description"]) refs = "" if e.get("references"): refs = " (" + ", ".join(f"#{r}" for r in e["references"]) + ")" lines.append(f"- {desc}{refs} by @{e['author']} in `{e['hash_short']}`") if e.get("breaking") and e.get("breaking_detail"): lines.append(f" > ⚠️ **BREAKING CHANGE:** {e['breaking_detail']}") lines.append("") ``` The plain-text generator also emits descriptions verbatim: ```python for e in entries[:5]: desc = (ai_rewrites or {}).get(e["hash_short"], e["description"]) lines.append(f" • {desc}") ``` ### Technical Analysis Commit descriptions, commit author names, breaking-change details, and `ai_rewrites` values can ...[truncated 2413 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (10)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding

The skill instructs the agent to read and write files and run shell commands, but it does not declare an explicit tool scope such as permissions or allowed-tools. That creates an authorization ambiguity where an agent may execute broader file or shell actions than the skill actually needs, increasing the chance of unsafe path access or command execution in real environments.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The scenario explicitly requires producing a 飞书 announcement, and the final output is written entirely in Chinese. This imposes a specific language/locale in the skill content without stating that the user can choose the language or that the Chinese output is optional or context-limited.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The prompt explicitly instructs the AI to rewrite all descriptions in concise Chinese and generate output for 飞书, which imposes a language requirement in the skill text. The file does not present this as an optional user-selected mode or a clearly justified region-specific policy constraint, so it constitutes a natural-language locale policy violation under the stated rules.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · examples/03-breaking-changes.md (reported line 91)May include surrounding context.

Change: Error responses now use RFC 7807 (Problem Details) format.

diff
- {"error": "Not Found", "code": 404}
+ {"type": "https://api.example.com/errors/not-found",
+  "title": "Resource Not Found",
+  "status": 404,
+  "detail": "User with id '99999' was not found"}

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The plain text section states support for platform variants, then shows --platform generic as the default using Chinese text (发布通知). This imposes a specific language/locale in the skill's natural-language behavior without opt-in or justification, which matches the language-policy violation criteria.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/weaver.py (reported line 84)May include surrounding context.

python
def run_git(repo_path: str, args: list) -> str:
    """Run a git command and return stdout."""
    try:
        result = subprocess.run(
            ["git", "-C", repo_path] + args,
            capture_output=True, text=True, timeout=30,
        )

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The plain-text generator hardcodes Chinese strings such as '发布通知' and '贡献者', and the formats help text advertises Chinese-only platform labels. This imposes a specific language/locale on users regardless of their preferences, which is a natural-language policy issue for a general-purpose script.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

This manifest description says the skill generates changelogs and release notes from common developer artifacts, but it does not define specific trigger phrases, scope boundaries, or exclusion conditions. In a manifest file, such broad wording can make invocation conditions ambiguous and increase the chance of unintended activation in ordinary development conversations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The sample prompt and behavior description explicitly require the output to be in Chinese for the DingTalk example. Under the policy, forcing a specific language is only acceptable when it is user-selected or clearly justified as region-specific; here the skill behavior is presented as doing Chinese output by default for that scenario without an explicit opt-in note or documented locale constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest sets the language to "en", which can indicate a fixed language preference without any user opt-in or explanation of why the skill must operate only in English. The provided file does not document a justified region-specific or compliance-driven reason for this locale constraint.

Content

No source excerpt is available for this finding.