The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests historical data from Claude Code session logs (~/.claude/history.jsonl) and project-specific message logs. Because this data originated from previous user interactions, it could theoretically contain instructions designed to manipulate the summary output.
Ingestion points: Reads from ~/.claude/history.jsonl and ~/.claude/projects/ directory files.
Boundary markers: The instructions do not employ specific delimiters to separate ingested history from the agent's internal instructions.
Capability inventory: The skill performs read-only operations on local history and git logs.
Sanitization: The collection script includes logic to redact common API key patterns (e.g., sk-, ghp-, ntn-) and skips strings that appear to be system commands or markdown tags.
[COMMAND_EXECUTION]: The skill executes a bundled shell script to perform its core data aggregation functions.
Evidence: The agent is instructed to run bash <skill-base-dir>/scripts/collect-daily-data.sh --date <argument>.
Context: The script is used locally to query git logs and session history; it does not involve network exfiltration or elevation of privileges.

daily-summary