feat(runtime): live observability (hooks passthrough) + the route-rigor auditor (auditIntent)#216
Merged
Merged
Conversation
…ent)
Online observability: RunAgenticOptions.hooks — the Supervisor lifecycle stream
(every spawn/settle: shots, analysts) now passes through runAgentic, so online
watchdogs/route-auditors/tree-views subscribe to a RUNNING loop, not just its report.
flywheel-run streams per-task lines on every phase (the onTask seam).
auditIntent — the route-rigor analyst, a different failure class from observe():
an agent can execute flawlessly down the WRONG ROUTE and no score sees it until
budget is gone. The auditor reads ONLY the trajectory + the intents (declared task /
user contract / loop meta-intent), infers the REVEALED intent from the action
pattern, and returns {verdict: aligned|drifting|diverged, evidence, recommendation:
continue|steer|abort, steer?}. Firewalled like every analyst (zero check-leakage —
always Goodhart-safe to run online). Runs between shots, as a stream watchdog
(abort-and-refund composes with the conserved pool: killing a doomed rollout buys
more attempts), or post-hoc over a BenchmarkReport (the meta-intent pass).
Verified live: a busy-but-wrong-route trajectory (task: escalate INC-7; agent:
cosmetic edits on unrelated tickets) -> diverged / abort / 0.98 confidence with
trajectory-grounded evidence.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two seams for the online layer: (1)
RunAgenticOptions.hooks— the Supervisor lifecycle stream passes throughrunAgenticso watchdogs/tree-views subscribe to RUNNING loops; flywheel streams per-task lines. (2)auditIntent— the route-rigor analyst: infers the trajectory's REVEALED intent and compares against declared/user/meta intents →aligned|drifting|diverged+continue|steer|abortwith evidence. Firewalled (trace-only ⇒ zero check-leakage ⇒ always safe online). Live-verified: caught a busy-but-wrong-route trajectory as diverged/abort @0.98 with grounded evidence.