The Agent Skills Directory

[COMMAND_EXECUTION]: The Preamble section contains instructions for the agent to execute a Bash script from a relative parent directory: bash "$(dirname "${BASH_SOURCE[0]}")/../../check-update.sh". Running scripts from outside the skill's defined path represents a dynamic execution risk.
[REMOTE_CODE_EXECUTION]: The skill's search strategy involves a component called 'AnySearch', which is described as locating a script and invoking it using python3. This indicates dynamic code generation and execution at runtime.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its data ingestion process.
Ingestion points: The skill reads existing files from docs/01-需求调研/市场调研报告.md and docs/01-需求调研/竞品监控月报.md (SKILL.md).
Boundary markers: None. The content is read and interpolated directly into subagent prompts (Step 2 and Step 4) without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has access to Bash, Write, and Task (subagent orchestration) tools, providing a path for injected instructions to perform unauthorized file operations or command execution.
Sanitization: There is no evidence of filtering, escaping, or validation of the content read from the file system before it is processed.

pm-competitor