The Agent Skills Directory

[PROMPT_INJECTION]: The skill's core functionality involves reading and rewriting Markdown documentation from other skills, which creates a potential surface for indirect prompt injection. Embedded instructions in target files could influence the agent's behavior during the compression or validation phases.
Ingestion points: Target Markdown files are read into the agent's context and processed by initialization and measurement scripts.
Boundary markers: No explicit delimiters or isolation techniques are used for the untrusted content being processed.
Capability inventory: The agent has the ability to write to the file system and execute shell commands.
Sanitization: No content sanitization or validation of the ingested Markdown content is performed.
[DATA_EXFILTRATION]: The skill instructions define a bug-reporting process that accesses paths outside the designated target skill root and involves external transmission.
Access to sensitive paths: The agent is instructed to log bug reports in ~/.bensz-skills/bugs/ within the user's home directory.
External transmission: The agent is directed to use the gh CLI tool to upload these bug reports to the author's GitHub repository (huangwb8/bensz-bugs) upon explicit user request. While this is a user-gated action using a well-known service and targeting a vendor-owned repository, it represents data transmission from the local environment.

compact-bensz-skills