Back to skill

Security audit

Agentsop Conventions Pinning

Security checks for vulnerabilities and agentic risk

Overview

The skill is a documentation-only guide for creating coding-agent style files, and its persistent behavior is disclosed and user-directed.

Before installing or using this skill, understand that it is meant to help you create files that influence future coding-agent sessions. Keep those files short, repo-scoped when possible, avoid secrets or private URLs, and review any user-global or always-loaded rules carefully because they can affect later agent behavior.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (6)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown file contains user-facing instructional text that assumes a mixed Chinese/English presentation, beginning with headings like "何时激活" and continuing throughout later sections. Under the stated policy, forcing a specific language or locale without opt-in is a natural-language policy concern unless the skill offers a choice or clearly justifies the locale constraint, which this file does not.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 379)May include surrounding context.

md
1. **Project narrative in CONVENTIONS.md**. "Our team started this in 2020 ..." costs tokens, produces zero behavior change. The research is clear: prose paragraphs *"reliably get ignored."* [blakecrosley.com/blog/agents-md-patterns]
2. **Duplicating what's in `package.json` / `pyproject.toml`**. The agent reads those. *"Context files generated by /init commands are doing little more than pre-caching information the agent would discover on its own."* [developer.upsun.com/posts/ai/agents-md-less-is-more] If it's already there, leave it out.
3. **Ambiguous directives**: "be careful with concurrency", "write good code", "follow best practices". These are observably ignored. [blakecrosley.com/blog/agents-md-patterns]
4. **Contradictory rules across user + project + ancestor**. Two CLAUDE.md files in the directory chain saying different things — Claude picks arbitrarily. Audit with `/memory`. [code.claude.com/docs/en/memory]
5. **TODO list inside conventions**. The file is *what the agent reads every session*, not your roadmap. TODOs cost tokens and the agent treats them as background noise.
6. **Pinning before stabilising**. Greenfield: write 2–4 modules first, then capture the choices.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This JSON manifest includes a trigger of "User asks how to make an agent respect project style," which is broad enough to overlap with ordinary support or advisory requests rather than a clear invocation boundary. The same field mixes several alternative conditions with no negative examples or explicit scope limits, making it unclear when this operation should activate versus when it should remain inactive.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The trigger includes phrases like "Same mistake twice" and "you re-typed the same correction in chat a third time," which rely on subjective interpretation and could match many routine interactions. Because the manifest does not define what counts as the same mistake or correction, the activation condition is underspecified and may lead to inconsistent or unintended invocation.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · references/R2-tool-equivalents.md (reported line 223)May include surrounding context.

md
## What none of these handle well

- **Cross-machine personal preferences**. CLAUDE.local.md is local to its directory; if you have 5 worktrees, you write it 5 times. Workaround: `@~/.claude/my-prefs.md` import in CLAUDE.md.
- **Truly cross-tool guarantees**. Each tool's adherence is probabilistic and tool-specific. The unification above gets you *one source of truth*, not *uniform behavior*.
- **Secrets / private URLs**. None of these are secret-safe (CONVENTIONS.md is committed). Use `CLAUDE.local.md` (gitignored) for personal sensitive paths.
- **Mutable working memory** — only Cline's `memory-bank/` has a built-in answer. For the others, working memory has to be reconstructed each session from the rules + the code.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

The manifest describes an SOP for project-level convention files such as CONVENTIONS.md, CLAUDE.md, .cursor/rules, .clinerules, and AGENTS.md, focused on persistent style guidance. At L418, the document introduces Cline's separate mutable working-memory feature (memory-bank/) and describes its behavior, which is a different capability than conventions pinning and is not necessary to fulfill the stated purpose.

Content

No source excerpt is available for this finding.