fix(sourcehunt): restore reliable reads and callgraph coverage - #190
Open
whatever wants to merge 16 commits into
Open
fix(sourcehunt): restore reliable reads and callgraph coverage#190whatever wants to merge 16 commits into
whatever wants to merge 16 commits into
Conversation
…stem The checkpoint system (#151-#164) resumes at stage granularity: a stage is restored only when it ran to completion, so a hunt interrupted part way through re-runs every already-finished file. PR #143 solved that with a separate work-item memoization path (its own resume.py, manifest, fingerprints, and --resume CLI), which collided with the checkpoint architecture. This folds #143's memoization *into* the checkpoint system instead of running a second one, so there is a single resume path with finer resolution. - hunt_work_cache.py: HuntWorkCache — an incremental, write-once per-work-item store (adapted from #143's resume.py load/save, minus the manifest / fingerprint / rank-plan machinery, which the checkpoint stages already provide). Malformed, missing, or non-completed entries are cache misses, so the work simply re-runs. - pool.py: WorkItem.stable_identifier folds in tier / entry-point identity / seed context / a per-file context_id so a cached result is never reused across changed inputs; HunterPool loads a completed result from the cache (as a resolved future) or runs and saves it. - findings_pool.py: FindingsPool.add is idempotent on finding id, so findings carried in from cached work are not double-counted on resume. - budget.py: SpendLedger(resume=True) replays the existing spend ledger so the dollar cap stays a lifetime budget across a resume; an interrupted reservation is charged once. - runner.py / CLI: the hunt stage always memoizes work items under <session>/hunt-work/; `--resume SESSION_ID` continues a prior session directory in place, routing entirely through the existing checkpoint / ledger / findings-pool machinery — no parallel resume path or CLI. Supersedes #143. Tests: 14 new (store, pool reuse, idempotent replay, ledger resume, work-item identity, --resume validation); full affected suite green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01ShCktZJoeBkVa3kGj6BKVS
Never put max_tokens, max_completion_tokens, or max_output_tokens on provider requests. Remove the rejection retry and cap-based truncation retry paths.
Co-authored-by: Matt Owen <matt.owen@lazarus.enterprises>
Co-authored-by: Matt Owen <matt.owen@lazarus.enterprises>
Co-authored-by: Matt Owen <matt.owen@lazarus.enterprises>
Co-authored-by: Matt Owen <matt.owen@lazarus.enterprises>
Run each finding through a bounded verifier tool loop in a fresh writable no-network sandbox. Record and redact all tool evidence, clamp evidence levels to host-observed crashes, fail closed on operational errors, and bind durable checkpoints to verifier inputs and policy versions. Co-authored-by: Matt Owen <matt.owen@lazarus.enterprises>
Bypass ranking for explicit repository-relative source files and seed each hunter with a bounded, line-numbered first-turn window. Bind target fingerprints, window ranges, and effective prompt context to checkpoints and granular work-cache identities, and fail incomplete plans closed before verification.
Contributor
|
are read behavior, callgraph, and otel provenace issues separable into different prs? |
Contributor
Author
I'll zap the provenance stuff since it's just about tracking performance for OTEL against commits. The callgraph stuff is just the preferred navigation mechanism, but is disabled for large code-bases (we should use a LSP approach as a follow-up) |
ehartford
force-pushed
the
codex/sourcehunt-07-target-file-windows
branch
8 times, most recently
from
August 30, 2026 07:23
7d5dfaf to
9b83919
Compare
Base automatically changed from
codex/sourcehunt-07-target-file-windows
to
main
August 30, 2026 07:32
Contributor
|
Please rebase @whatever |
This was referenced Sep 1, 2026
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
3 participants
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.
Summary
read_filecoverage using the lines actually returned, including requested/returned ranges, EOF, truncation, overlap, and new-line metadataValidation
git diff --checkpassedStack and integration
codex/sourcehunt-07-target-file-windows)Local state
Unrelated local evaluation-ledger and skill files were excluded from this PR.