The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled shell script scripts/collect-daily-data.sh which further spawns a Python subprocess using a heredoc to parse and process session logs.
[DATA_EXFILTRATION]: The skill accesses and exposes sensitive local data to the LLM context to perform its summary task:
Accesses Claude Code session history at ~/.claude/history.jsonl and individual project session logs.
Aggregates Git commit messages from various repositories, including a hardcoded path to the user's ~/dotfiles directory, which often contains private configurations.
Note: The skill implements a security feature to redact common API tokens (e.g., ntn_, sk-, ghp_) from the session content before it is passed to the agent.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests untrusted data from Git commit messages and previous Claude sessions:
Ingestion points: ~/.claude/history.jsonl, ~/.claude/projects/**/*.jsonl, and local .git logs.
Boundary markers: Absent; the script output does not use delimiters to isolate untrusted content from instructions.
Capability inventory: The agent uses the aggregated data to generate a natural language summary.
Sanitization: Includes regex-based redaction of secrets, but does not sanitize potential instruction-like text in commit messages.

daily-summary