The Agent Skills Directory

[COMMAND_EXECUTION]: The installation process (install.sh) modifies the agent's global configuration files (~/.claude/settings.json and ~/.codex/hooks.json) to register a persistent SessionStart hook. This hook executes a shell script (obs-orient.sh) every time a session begins, ensuring the agent's context is automatically populated with vault state.
[DATA_EXFILTRATION]: The analyze command is designed to systematically harvest extensive information from the current repository, including remote repository URLs (via git remote), directory trees, and the full content of various configuration and documentation files (e.g., CLAUDE.md, .cursorrules, package.json). This data is aggregated into a centralized Obsidian vault.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection during the repository analysis phase. It ingests untrusted data from documentation (README.md, ADRs) and agent configuration files without explicit sanitization or strict boundary markers. Malicious instructions placed in these repository files could influence the agent's behavior when the analyze command is executed.
[COMMAND_EXECUTION]: The skill relies on and executes the third-party obsidian CLI tool and several system commands (git, mkdir, cp, jq, sed) to manage the knowledge graph and project metadata, as declared in the allowed-tools configuration.

obs-memory