Skip to content

Local-first 🧠 memory for Hermes Agent that lives in your Obsidian vault and remembers project context. Nightly 😴 dream passes turn repeat corrections into confirmed preferences with measurable confidence. Adapters ship for Claude Code, Codex, and OpenClaw, with an MCP server for anything else.

TypeScript
1.2 KB
Raw Blame

ADR: Agent capability + CLI integration foundation

Status

Accepted for implementation in this feature branch.

Context

The board comments for t_4acb2c72 note that runtime tool capability verification needs an ADR before implementation because Open Second Brain currently has only static MCP ToolScope filtering. The same PR also includes two CLI integration tasks: inherited --json output and shell completions.

Decision

Add two separate foundations:

  • MCP runtime capabilities remain a small layer above static ToolScope. They can withhold tools only after static scope has already selected the candidate set, and every withheld tool must appear in a diagnostic report with a reason.
  • CLI discovery/completions use a CLI-only command manifest. This manifest does not own command dispatch; it owns machine-readable metadata for help JSON, fallback JSON, and shell completion generation.

Consequences

  • Static writer scope remains the access-control baseline.
  • Agents can ask why a tool is unavailable instead of inferring from a missing tool list.
  • CLI completions and discovery share metadata without a new CLI framework dependency.
  • The project keeps two intentionally small registries rather than one large cross-surface registry.