The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses git commands to identify the project directory and curl to fetch observation data from a local API running on localhost:37777. It also instructs a subagent to execute sqlite3 commands against the local database file ~/.claude-mem/claude-mem.db to calculate development metrics.
[COMMAND_EXECUTION]: The skill accesses the local claude-mem database, which contains sensitive historical data including previous prompts, file modifications, and project observations. This access is required for the skill's primary purpose of summarizing development history.
[PROMPT_INJECTION]: The skill processes potentially large volumes of historical 'observations' which could contain adversarial instructions (Indirect Prompt Injection). A subagent is used to analyze this untrusted content to generate the final report.
Ingestion points: Historical data is ingested from a local API and an SQLite database (SKILL.md).
Boundary markers: The prompt for the subagent includes the project timeline at the end of the instructions but lacks robust delimiters or sanitization to prevent the agent from obeying instructions embedded in the historical data.
Capability inventory: The subagent can read from the local file system via sqlite3 and produce markdown reports.
Sanitization: No sanitization or escaping of the timeline content is performed before it is analyzed by the subagent.

timeline-report