The Agent Skills Directory

Indirect Prompt Injection Surface: The skill reads communication logs from local storage to generate summaries and identify patterns. Since these logs represent summaries of external communications, they could theoretically serve as a vector for adversarial instructions. However, the skill's capabilities are limited to local file operations, and the required user confirmation for new entries provides a significant human-in-the-loop safeguard. \n
Ingestion points: Data is read from ~/.claude/plugins/config/claude-for-legal/legal-clinic/client-comms/[case-id]/log.md (SKILL.md) during the --read, --summary, and --patterns operations.\n
Boundary markers: While the logs are structured as Markdown entries, no specific delimiters are used to isolate potentially untrusted communication text from the agent's instructions.\n
Capability inventory: The skill is restricted to reading and appending to Markdown files within a specific configuration path; it has no network or shell execution capabilities.\n
Sanitization: The skill relies on manual user review and confirmation before any new data is written to the log, which acts as a filter for problematic content.

client-comms-log