The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from potentially untrusted external sources, including GitHub pull request titles and commits, Jira tickets, and local Claude Code session history. This creates an attack surface where malicious instructions embedded in these external sources could theoretically influence the agent's behavior during the interview or summary generation process.
Ingestion points: Data is pulled from GitHub and Jira via their respective CLIs and from local Claude Code session logs (~/.claude/projects/*.jsonl) via a Python script.
Boundary markers: The instructions in SKILL.md do not specify the use of delimiters or 'ignore' instructions when presenting this retrieved data to the model.
Capability inventory: The skill possesses the capability to execute shell commands (gh, jira, python3) and read files from the user's home directory.
Sanitization: Neither the SKILL.md instructions nor the claude_digest.py script implement sanitization or filtering to prevent instruction injection from the ingested content.
[COMMAND_EXECUTION]: The skill executes local shell commands to detect the presence of integrations and to run a Python-based digestion script. While these are functional requirements for the skill, the reliance on shell execution is an inherent capability that should be noted.
Evidence: SKILL.md describes executing gh auth status, jira, and python3 ~/.claude/skills/daily-meeting-update/scripts/claude_digest.py to facilitate data retrieval.

daily-meeting-update