The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local command-line interface tools to reconstruct work history. It specifically instructions the agent to execute git log to retrieve commit messages and gh pr list to fetch pull request metadata from the GitHub API. These actions are aligned with the stated purpose of the skill to identify accomplishments across repositories.
[INDIRECT_PROMPT_INJECTION]: The skill ingests text from external, potentially untrusted sources which could contain malicious instructions or deceptive content aimed at the LLM.
Ingestion points: Commit messages retrieved via git history, pull request titles and descriptions from the GitHub CLI, and workspace summaries stored in ~/.copilot/session-state/workspace.yaml.
Boundary markers: The skill lacks explicit instructions for the agent to use delimiters or ignore embedded instructions within the processed text, increasing the risk of the agent following instructions found in logs.
Capability inventory: The agent has permissions to read local file metadata, execute shell commands (git/gh), and interact with the save_to_brag_sheet tool if present.
Sanitization: The primary mitigation provided is a manual review requirement, where the agent is instructed to 'show drafted entries to the user before saving'. This prevents silent exploitation but does not involve automated sanitization of the input data.

brag-sheet