Hackpert: re-evaluate objective on each loop step #80

Closed
nsaspy wants to merge 2 commits from rage-hackpert-12-objective-loop-step into master
Owner

Slice

Advance #29/#27 with one pure Hackpert control-plane boundary that re-evaluates objective evidence and extension applicability on every loop iteration before deriving the next symbolic/direct decision.

This does not execute providers, mutate canonical state, own persistence, or add another scheduler/executor. Authority remains orthogonal to reasoning strategy and effects still cross the existing canonical Hackmode boundary.

RED

Test-only head 4d12338f76f6b33d8d8143058e8c530445b00a71 failed core exactly because HACKMODE:EXPERT-OBJECTIVE-LOOP-STEP did not exist. Monorepo and agent-framework-boundary passed.

GREEN

Exact current head d0ff9b626391e6ada768d45c066a54f9765b84c5 passed fresh core, monorepo, and agent-framework-boundary on draft PR #119. This non-draft replacement uses the identical head because GitHub's ready-for-review mutation failed on the recurring fullDatabaseId GraphQL connector bug.

The loop step validates plan/objective identity, re-runs objective evaluation and extension admission using the current reasoning strategy, derives goal/no-extension stop signals from fresh results, and delegates the final transition to the existing generic loop.

No database-owned files or StarIntel product code are touched.

Issue: #29

## Slice Advance #29/#27 with one pure Hackpert control-plane boundary that re-evaluates objective evidence and extension applicability on every loop iteration before deriving the next symbolic/direct decision. This does not execute providers, mutate canonical state, own persistence, or add another scheduler/executor. Authority remains orthogonal to reasoning strategy and effects still cross the existing canonical Hackmode boundary. ## RED Test-only head `4d12338f76f6b33d8d8143058e8c530445b00a71` failed core exactly because `HACKMODE:EXPERT-OBJECTIVE-LOOP-STEP` did not exist. Monorepo and agent-framework-boundary passed. ## GREEN Exact current head `d0ff9b626391e6ada768d45c066a54f9765b84c5` passed fresh core, monorepo, and agent-framework-boundary on draft PR #119. This non-draft replacement uses the identical head because GitHub's ready-for-review mutation failed on the recurring `fullDatabaseId` GraphQL connector bug. The loop step validates plan/objective identity, re-runs objective evaluation and extension admission using the current reasoning strategy, derives goal/no-extension stop signals from fresh results, and delegates the final transition to the existing generic loop. No database-owned files or StarIntel product code are touched. Issue: #29
feat: re-evaluate objective on each expert loop step
Some checks failed
agent-pr-serialization / one-open-pr-per-agent-lane (pull_request_target) Failing after 1s
agent-framework-boundary / keep-agent-framework-out-of-product-tree (pull_request) Successful in 8s
core / common-lisp-core (pull_request) Failing after 51s
monorepo / hygiene (pull_request) Successful in 4m4s
d0ff9b6263
nsaspy closed this pull request 2026-09-04 20:55:35 +00:00
Some checks failed
agent-pr-serialization / one-open-pr-per-agent-lane (pull_request_target) Failing after 1s
agent-framework-boundary / keep-agent-framework-out-of-product-tree (pull_request) Successful in 8s
core / common-lisp-core (pull_request) Failing after 51s
monorepo / hygiene (pull_request) Successful in 4m4s

Pull request closed

Sign in to join this conversation.
No description provided.