Add scoped navigation, worker packets and dogfood cost reporting - #12
Open
TheCryptoDonkey wants to merge 8 commits into
Open
TheCryptoDonkey wants to merge 8 commits into
TheCryptoDonkey wants to merge 8 commits into
Conversation
TheCryptoDonkey
marked this pull request as ready for review
September 21, 2026 11:42
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Repeated source discovery and broad worker prompts make internal Context adoption difficult to assess. This change adds explicitly scoped repository navigation, verified worker source packets and a developer report for whole-task cost evidence.
Navigation honours root/nested Git ignores and literal include/exclude policy. Generations bind to observed root identity; changed or unverifiable policy blocks old results until refresh. Policy is rechecked after source reads and before cursor commit, source handles must match discovered files, and cancellation preserves cursor semantics. These are local project-selection controls, not an OS sandbox or signed-room grant.
The packet helper retains exact excerpts, whole-file hashes, allowed-file states, Git HEAD and policy provenance. It bounds input/output, creates private files exclusively and verifies freshness by rebuilding against an explicit root. It does not invoke models or execute acceptance commands.
The new cost reporter checks eight-pair protocol and receipt declarations, rejects mismatched controls, preserves unknown costs and coverage, includes failed attempts and keeps cached/reasoning usage separate. Cost and review decisions use exact decimal comparisons. It makes no provider calls and neither authenticates billing evidence nor estimates prices. The included template is explicitly unlocked with no real results. M4 worker routing and actual failed drafts, repairs and review are recorded in the execution ledger.
Validation: 170 tests (33 core, 107 tools, 14 packet, 16 cost-report), independent packed-package/browser/CLI smoke, and both unchanged benchmark gates. Six independent synthetic report CLI cases cover exact thresholds, unknown evidence, quality and review failures, and an empty draft. Sol/high reviewed the reporter after accounting repairs. Existing pinned navigation acceptance covers the nine-check two-process smoke, actual Codex edit/stale/refresh/old-cursor lifecycle and reciprocal scoped KithMoot/Oathrun retrieval; this developer-only script does not change that installed runtime.
Remaining gates: disabled-session tool inventory lacks raw client evidence; the eight-pair whole-task cost experiment, consumer room integration and public registry release remain open. No monetary savings are claimed. See docs/DOGFOOD-EXECUTION.md and docs/TASK-COST-REPORT.md for evidence and limitations.