The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell commands in SKILL.md to compute a SHA256 checksum of its reference knowledge pack (references/pm-knowledge-pack.md) to ensure instruction integrity before processing. This is a security best practice for ensuring the agent uses the correct version of its shared knowledge.
[DATA_EXPOSURE_EXFILTRATION]: The skill references a local configuration file at ~/.config/zynkr/pm.json to retrieve project-specific identifiers (Google Sheet/Doc IDs). This follows standard security practices for decoupling secrets/identifiers from the skill code. No evidence of unauthorized network exfiltration was found.
[DYNAMIC_EXECUTION]: The provided Python script scripts/pm-schema.py performs static analysis of other code files using the ast (Abstract Syntax Tree) module. This is used for validating schema and constants without executing the target files directly, which is a safe approach to dynamic configuration management.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied meeting transcripts and notes. While this is a common attack surface, the skill includes explicit instructions to 'only organize, do not invent' (只整理、不虛構) and provides strict structural boundaries for the output, mitigating the risk of embedded instructions in the source text influencing agent behavior.

project-note-specialist