The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions specify reading the .git/config file to determine the repository's remote origin. This file is sensitive as it can contain authentication tokens or private repository metadata. The access is used to establish project identity for session continuity.
[COMMAND_EXECUTION]: The agent is directed to execute git log and gh api commands. These are used to gather local commit history and remote progress updates from GitHub issue comments to reconstruct session state.
[EXTERNAL_DOWNLOADS]: Fetches data from GitHub's official API using the gh command-line tool. This is used to read recent comments from specified repository issues. As GitHub is a well-known service, this is a standard operation for development context recovery.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting content from GitHub comments and local phase-tracking files without sanitization or boundary markers.
Ingestion points: Data enters the session briefing from .harness/state.json, .harness/phases/, MEMORY.md, and GitHub issue comments via gh api.
Boundary markers: None; the external content is directly included in the structured briefing output provided to the agent.
Capability inventory: The skill uses file system reads, git log, gh api, and the roam_retrieve tool.
Sanitization: No validation or escaping is applied to the ingested external content before it is processed.

session-start