The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md define shell commands that incorporate template placeholders such as {slug}, {self_content}, and {persona_content} without explicit validation or escaping directives. For example, the command rm -rf .claude/skills/{slug} and the cat heredoc in the file-writing step could be exploited if a malicious value is substituted into these variables, leading to unauthorized file deletion or arbitrary code execution within the shell environment.
[PROMPT_INJECTION]: The skill employs 'identity override' instructions, telling the agent 'You ARE {name}, not an AI assistant' and providing a strict set of 'Hard Rules' (Layer 0) to bypass default AI behavior. While functional for the skill's purpose, these are standard prompt injection techniques used to subvert core system instructions.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted external data from chat logs and social media via tools like wechat_parser.py and qq_parser.py. This data is then analyzed by the agent (e.g., in prompts/self_analyzer.md) without the use of sanitization or boundary markers (delimiters) to distinguish between data and instructions.
Ingestion points: Raw text extraction from files via SKILL.md (Step 2), wechat_parser.py, qq_parser.py, and social_parser.py.
Boundary markers: None. The analysis prompts in the prompts/ directory do not utilize delimiters or specific instructions to ignore embedded commands.
Capability inventory: The skill is configured to use Bash, Write, Edit, and Read tools, providing a significant attack surface if an injection occurs.
Sanitization: None. The parsing scripts extract raw text and metadata without filtering or escaping.
[EXTERNAL_DOWNLOADS]: The requirements.txt file specifies external Python dependencies including requests, Pillow, and pypinyin. While these are established libraries, they introduce a dependency on external registries and code not contained within the skill itself.

yourself-skill