The Agent Skills Directory

[COMMAND_EXECUTION]: Uses the Bash tool to execute environment audits and maintenance tasks.
Evidence: Executes git status, git log, and wc for environment checks. It also invokes project-local scripts such as scripts/harness-memd for recovery.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its reliance on reading and processing external data files.
Ingestion points: Reads task status and constraints from Plans.md, AGENTS.md, and session logs.
Boundary markers: No delimiters are used to isolate content from instructions.
Capability inventory: The skill utilizes Bash, Read, and Write permissions.
Sanitization: Content from the file system is processed without validation or filtering.
[DATA_EXFILTRATION]: The skill accesses sensitive local files to restore session context.
Evidence: Reads from a session database at ~/.harness-mem/harness-mem.db and session metadata from .claude/state/session.json.

session-init