The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses local files and history logs that typically contain sensitive information.
Scans Zsh history and files changed today, which may contain sensitive commands, arguments, or environment variables.
Accesses local Git repositories under ~/ghq and reads commit history and diffs.
Processes LLM transcripts and history from Codex and Claude.
Note: The skill includes developer-provided instructions to filter credentials and omit secrets from the output, which mitigates the risk of accidental exposure during the summary process.
[COMMAND_EXECUTION]: The skill relies on executing local system tools to gather information and interact with the environment.
Uses gh to retrieve GitHub activity.
Uses git (implied by ghq and diffing instructions) to inspect local repository state.
Uses pbcopy to write the final summary to the macOS system clipboard.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted data that could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Reads GitHub PR descriptions, Git commit messages, and shell history entries (SKILL.md).
Boundary markers: The instructions mention filtering but do not define explicit structural delimiters or "ignore" warnings for the ingested text.
Capability inventory: The agent has the capability to execute shell commands (gh, pbcopy) and read local file system contents (SKILL.md).
Sanitization: The skill instructions require filtering for credentials, but do not specify sanitization or escaping for the content of processed commit messages or PR descriptions to prevent prompt injection.

telegram-daily-log