Tier-1 Instant Answers (tier1)

Configures Tier-1 instant answers — a narrow, exact-match command set answered deterministically by the prompt_tier1 op with zero model invocation. The prompt never reaches Claude: the hook runtime matches it, gathers the data locally, and renders a terminal card. Zero tokens spent, instant response.

New in v7.0.0. Ships off (new blocking features seed off); you opt in explicitly.

Config block

In .agent/.nav-config.json:


{
  "tier1": {
    "enabled": false,
    "rules": {}
  }
}

Keys

enabled (default false) — Master switch. When true, the prompt_tier1 op intercepts matching prompts on UserPromptSubmit and answers them directly.
rules (default {}) — Per-rule toggles keyed by rule name. Set a rule to false to let that command fall through to the model while others stay instant. Omitted rules follow the master switch.

The five commands

Only these exact-match commands are handled; anything else goes to the model as usual:

Command	Answer
`nav stats`	Session efficiency report (terminal card).
`show features`	Current feature toggles from `.nav-config.json`.
`list markers`	Context markers on disk.
`graph health`	Knowledge-graph health summary.
`nav version`	Installed Navigator version.

Matching is exact — nav stats please or can you show features are not intercepted. This keeps false positives at zero: if there is any ambiguity, the model handles it.

Enabling

Globally:


{
  "tier1": { "enabled": true }
}

Per rule — enable Tier-1 but keep one command routed to the model:


{
  "tier1": {
    "enabled": true,
    "rules": { "nav_stats": false }
  }
}

Escaping to the model

Every Tier-1 card carries the escape hatch in its footer:


reply 'ask claude' to run the model

If you want Claude’s interpretation of an intercepted command (for example, stats with commentary), reply “ask claude” after the card — that prompt is not on the whitelist, so it reaches the model normally. Any non-exact rephrase (ask claude nav stats, can you show features) also falls through to the model.

Interaction with PILOT_EXECUTOR

Setting the PILOT_EXECUTOR environment variable disables interactive/blocking hook behavior across all ops, including Tier-1 interception, regardless of config.