agentfootprint
Version:
The explainable agent framework — backtrack a wrong answer to the exact context that caused it (evidence, not guesses). Built on footprintjs.
55 lines • 2.49 kB
JavaScript
/**
* traceToolNames — the two facts about the trace toolpack that are needed
* BEFORE the pack is loaded.
*
* `.selfExplain()` mounts the pack lazily (9.94.0): the toolpack module —
* eleven tools over a finished trace, the largest single module on the
* library's default graph — is reached through `import()` the first time
* the self-explain skill is active on an iteration, never at build time and
* never for an agent that did not enable it (see `selfExplain.ts` ·
* `buildSelfExplainToolProvider`). Two things the builder needs cannot wait
* for that load, and they live here so that reading them costs nothing:
*
* - `TRACE_TOOL_NAMES` — the names the pack can mount, which `build()`
* reserves synchronously (a consumer tool sharing a name would silently
* shadow the trace tool);
* - `NO_COMPLETED_RUN_MESSAGE` — the model-visible answer every trace tool
* gives before a turn has finished, which the delegate-mode `explain_run`
* tool returns without loading the pack at all.
*
* `traceToolpack.ts` re-exports `TRACE_TOOL_NAMES` and `lazyToolpack.ts`
* re-exports `NO_COMPLETED_RUN_MESSAGE`, so every existing import path still
* resolves — this module only changes what a static import of them COSTS.
*/
/**
* The tool names this pack can mount. The Agent builder reserves these at
* `.selfExplain()` build time — the tools slot dedupes by name with
* first-occurrence-wins, so a consumer tool sharing a name would silently
* shadow the trace tool AND the skill body would instruct the model into
* the wrong one. Exported so the reservation list is DERIVED from the pack
* rather than typed out beside it (a second list is a list that drifts).
*
* The pack itself asserts, at test time, that what it mounts is exactly
* this list (test/trace-toolpack/selfExplainAgent.test.ts).
*/
export const TRACE_TOOL_NAMES = [
'run_overview',
'find_context_errors',
'find_in_trace',
'trace_node',
'trace_slice',
'backtrack',
'who_wrote',
'get_value',
'inspect_tool_call',
'inspect_tool_run',
'inspect_subflow',
'attention_list',
'root_cause',
'impact',
'read_narrative',
];
/** Model-visible answer when no completed run is available yet. */
export const NO_COMPLETED_RUN_MESSAGE = 'No completed run is available yet — the trace exists only after a turn finishes. ' +
'Tell the user there is nothing to explain yet.';
//# sourceMappingURL=traceToolNames.js.map