Session Tools

session_init

Single-call session bootstrap. Returns pending tasks, working state, and scale guidance in one round-trip. Supports incremental delivery when agent_id is provided across calls.

Absorbs: explain_codebase, get_repo_map, get_project_identity (via scope=“full”), set_sdlc_phase, set_quality_mode.

Parameters

Parameter	Type	Required	Description
`agent_id`	string	No	Self-declared agent identifier. Enables incremental delivery: subsequent calls skip unchanged sections. Always provide for token savings.
`model`	string	No	Model name (e.g. `claude-sonnet-4-6`). Recorded in pulse analytics for cost calculations.
`provider`	string	No	Model provider: `anthropic`, `openai`, etc.
`intent`	string	No	Short free-text declaration of current work. Visible to peer agents. Pass empty string to clear.
`scope`	string	No	Controls response verbosity. `standard` (default): tasks + working_state + scale_guidance (~500 tokens). `full`: all sections including project_identity, brain_health, federation_health, relevant_memories, knowledge_graph. `resume`: tasks with session states + working_state + relevant_memories + federation_health. `quick`: alias for standard.

Example

{
  "tool": "session_init",
  "arguments": {
    "agent_id": "claude-1",
    "scope": "standard"
  }
}

Response Sections by Scope

Section	standard	full	resume
pending_tasks	yes	yes	yes
working_state	yes	yes	yes
scale_guidance	yes	yes	yes
more_available	yes	no	no
project_identity	no	yes	no
brain_health	no	yes	no
federation_health	no	yes	yes
relevant_memories	no	yes	yes
knowledge_graph	no	yes	no

Safety-critical alerts (cross_project_alerts, agent_awareness, tool_integrity_alert) are always included regardless of scope.

end_session

Captures session knowledge and persists it as structured memories. Call at the end of every session. Optionally reports LLM token usage (absorbs report_usage).

Absorbs: report_usage.

Parameters

Parameter	Type	Required	Description
`agent_id`	string	Yes	Self-declared agent identifier.
`task_id`	string	No	Link session memories to this task.
`summary`	string	No	High-level summary of what was accomplished. Saved as a project-tier memory.
`model`	string	No	Model name for usage reporting (e.g. `claude-sonnet-4-6`).
`provider`	string	No	Model provider. Used with `model` for usage reporting.
`input_tokens`	number	No	Total input tokens consumed during this session.
`output_tokens`	number	No	Total output tokens generated during this session.
`cost_usd`	number	No	Total USD cost for this session if known.

Example

{
  "tool": "end_session",
  "arguments": {
    "agent_id": "claude-1",
    "summary": "Implemented user authentication flow",
    "model": "claude-sonnet-4-6",
    "provider": "anthropic",
    "input_tokens": 45000,
    "output_tokens": 12000
  }
}

Response

Returns an effectiveness_report with session quality metrics:

context_hit_rate — fraction of context calls that were useful
first_fetch_right / total_deliveries — how many context calls required no correction
tokens_saved — estimated token savings from incremental delivery
prev_7d — comparison to the last 7 days of sessions
message — human-readable session summary