Skip to content

Latest commit

 

History

History
41 lines (33 loc) · 2.07 KB

README.md

File metadata and controls

41 lines (33 loc) · 2.07 KB

terraphim-clients

Client + integration crates extracted from terraphim-ai (#1910):

  • terraphim-cli / terraphim_agent -- CLI + agent REPL
  • terraphim_mcp_server -- MCP server
  • terraphim_sessions -- session import/search
  • terraphim_lsp -- LSP server (EDM diagnostics)
  • terraphim_grep, terraphim_hooks, terraphim_update, terraphim_command_runtime, terraphim_negative_contribution

Consumes upstream crates (core, config-persistence, service, agents, kg-agents) from the terraphim Gitea cargo registry. Licensed Apache-2.0.

Memory Lifecycle

The terraphim-agent memory namespace saves authored lessons, inspects the evolution store and retrieves memories through a configured role's knowledge graph. See capture and store isolation for a complete example and the required input change.

terraphim-agent memory capture --content 'Use the project validation command.'
terraphim-agent memory scope       # Show role/project KG boundaries
terraphim-agent memory retrieve    # Search memory items
terraphim-agent memory apply       # Show hook injection effects
terraphim-agent memory validate    # Score items against reliability rubric
terraphim-agent memory retire      # Propose demotion of stale items
terraphim-agent memory rubric      # Full 6-dimension diagnostic
terraphim-agent memory second-run  # Token delta between ADF runs
terraphim-agent memory list        # Browse evolution store
terraphim-agent memory show        # Inspect a specific item
terraphim-agent memory export      # Dump as JSON or markdown

The full eight-stage lifecycle is not qualified by these commands. In the current implementation, memory distill and memory provenance only report status; memory scope --check does not verify confidentiality. Use explicit learn compile / learn export-kg for the separate reviewed-learning store. memory apply previews context; it does not execute a command. Rubric and validation results use the named heuristic-v1 scorer.

See MEMORY_POLICY.md for the public commons vs permissioned memory boundary.