Validate

validate answers one question: is this context enough to act on? It's the guardrail between retrieval and action.

Basic use

tokengraph validate "auth login token"

It retrieves a pack for the task and reports a coverage score.

Set a threshold (CI / hook gate)

tokengraph validate "auth login token" --min-coverage 60

Exits non-zero when coverage falls below the threshold — so an agent or a CI step can refuse to proceed on thin context instead of guessing.

Flags

Flag	Meaning
-b, --budget	Token budget for the trial pack
-d, --depth	Graph traversal depth
--min-coverage	Fail below this coverage percentage (0–100, default 60)
--json	Machine-readable output

Where it fits

ask → validate → (AI answers) → judge → verify
        ▲
   stop here if coverage is low —
   refine the task or widen the budget

If coverage is low, the usual fixes are: make the task string more specific, raise --budget, or increase --depth.

Next steps

• After the answer comes back: Judge and Verify
• Wire it into a PreToolUse hook — see MCP server