cogsession
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| session_initA | Initialize a new CogSession for the current project. Call this at the START of a Claude Code session. Provide parent_session_id to continue from a previous session (tree structure). Provide focus to describe what this session is about. |
| session_checkpointA | Save the current session state to disk. CALL THIS at 70-80% context. Writes: handoff.md, tasks.json, dead_ends.md, assumptions.md, environment.json, architecture.mermaid, session_log.jsonl. Also auto-writes handoff.md to CLAUDE.md so next session loads it. |
| session_loadA | Load the handoff from a previous session. Call at the START of a new session to get context from the last one. Returns the handoff brief (~250 tokens) — exactly what the new session needs. |
| session_updateB | Add information to the current session. Use this throughout the session to record: decisions made, dead ends found, assumptions made, tasks completed/added, danger zones discovered, errors seen. |
| session_treeB | Show the full session tree for this project. Like 'git log --graph' but for CogSession work history. Shows all sessions, their relationships, and their status. |
| session_searchA | Search across ALL sessions for a query. Useful for: 'when did we decide X?', 'what errors have we seen?', 'which session touched file Y?', 'what dead ends are there?' |
| session_statusA | Get the current session's status: what's been recorded, how many dead ends, task progress, context warning level. |
| session_logA | A |
| claim_recordA | Record a claim together with the command that PROVES it, so a later session is told when it stops being true. Use this for any factual statement you write somewhere durable — a PR description, a code comment, a doc, a status report. The proof must be a cheap read-only shell command whose output can be compared. |
| claim_checkA | Re-run recorded claims and report only the ones that no longer hold. Silence means everything still checks out. |
| session_diagramB | Show or regenerate the architecture diagram (Mermaid format). Auto-scans the project files to build a dependency graph. Pass to a Mermaid renderer to visualize. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 11 tools
Most tools are clearly distinct — checkpoint/update, tree/search/log/status, and claim_record/claim_check each serve different purposes. The main ambiguity is session_init and session_load, since both are meant to be called at session start and both reference continuation from previous sessions, though their outputs differ.
The session_ prefix gives the set a strong, consistent identity, and all names use lowercase_with_underscores. However, operations mix imperative verbs (init, load, update) with noun-style view commands (tree, status, log, diagram), and the claim_* pair breaks the session_ prefix pattern.
11 tools is well-scoped for a session-management server. Each tool covers a distinct part of the workflow — lifecycle, history, search, status, claims, and diagram — and none feels redundant or unnecessary.
The core lifecycle is well covered: init, update, checkpoint, load, plus history, search, status, claims, and diagram. Minor gaps exist — there is no explicit session_end/archive tool and no way to delete or supersede a claim — but agents can complete the main workflow without dead ends.