Marius CiepluchandClaude Opus 5.5 3ffb8b4c54 fix(hunter): ask for the final answer in plain text; DeepSeek's JSON mode lost it ...

The final-answer turn requested the provider's JSON mode with the tools still
attached. DeepSeek (deepseek-flash) then answered with the format itself
(`{"type": "json_object"}`), wrapped the array (`{"type": "json_object",
"value": [...]}`), or emitted its tool-call markup as text. A hunt that had found
the vulnerability reported nothing.

Measured on the v2 sink verifier with the declared vulnerable functions: 0 of 7
confirmed in JSON mode, 5 of 7 asked plainly (6 of 7 at 8 steps). The model named
Medplum's actual redirect-URI prefix-match bug, LMDeploy's redirect bypass of
_is_safe_url, and ContextForge's unsandboxed Jinja2.

The final turn is now plain text without tools. The parser also reads a
`{"value": [...]}` wrapper. Two tests that pinned JSON mode now pin the plain
turn, and a new test reproduces the wrapped answer.

Invalidates as measurements: the hunter experiment on v2 (c4a0bd8) and the first
sink-verifier run. Both used this turn.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_01EmM1j46qgKMh3zqD1QMwEo

2026-09-28 20:10:57 +02:00

..

golden

feat(gate): Phase 4 task 8 — lock the detection gate into CI

2026-06-22 17:13:02 +02:00

conftest.py

fix(cpg): the real-request cost, attributed and removed -- overlays once, seeds scoped, data tables out

2026-09-11 01:16:00 +02:00

test_benchmark.py

feat(three-stage-scan): add aliased-sink mutation the regex scan misses

2026-09-04 11:50:44 +02:00

test_calibration.py

ci: add quality gates and release workflows

2026-06-18 10:06:21 +02:00

test_candidate_enumeration.py

feat(model-grounded-detection): group 1, subtract the noise architecture

2026-09-08 12:15:54 +02:00

test_cli_hx_dispatch.py

feat(contributor-scan): 1.3 the model layer reaches a user

2026-09-09 07:42:44 +02:00

test_cli_improve.py

feat(pair-corpus-honesty): reject improve rounds that regress any provenance profile on holdout

2026-09-05 10:45:43 +02:00

test_complexity_hints.py

feat(three-stage-scan): attach test-gap hints to hotspots

2026-09-03 20:33:48 +02:00

test_complexity_ledger.py

feat(three-stage-scan): report map and honor fail-on worth-fixing

2026-09-04 00:18:33 +02:00

test_complexity_map.py

feat(three-stage-scan): attach test-gap hints to hotspots

2026-09-03 20:33:48 +02:00

test_complexity_signals.py

feat(three-stage-scan): compute complexity signals for hotspot ranking

2026-09-03 20:05:13 +02:00

test_config.py

feat: load OpenRouter settings from a local .env

2026-09-04 14:06:18 +02:00

test_cpg_backend.py

fix(scan): report requests the reserve held back as unasked, not failed

2026-09-23 08:51:12 +02:00

test_cpg_budget.py

test(cpg-budget): a process vanishing mid-read is the success case too

2026-09-26 07:44:33 +02:00

test_cpg_queries_compile.py

Revert "feat(taint): project sink wrappers, one level and confirmed by dataflow"

2026-09-26 15:36:25 +02:00

test_cpg_session.py

fix(cpg): stop a failed session start from consuming the transaction budget

2026-09-17 08:35:57 +02:00

test_degradation_matrix.py

feat(model-grounded-detection): group 1, subtract the noise architecture

2026-09-08 12:15:54 +02:00

test_detection_benchmarks.py

feat(gate): Phase 4 task 8 — lock the detection gate into CI

2026-06-22 17:13:02 +02:00

test_dispatch_value.py

fix(mapping): record every callback of a register_rest_route call as a route

2026-09-24 11:08:22 +02:00

test_fact_memo.py

perf(push): find the Python-side scale walls all at once, and remove them

2026-09-23 19:36:32 +02:00

test_findings.py

feat(ruleset): Phase 2 (task 3) — rules as governed data, severity from policy

2026-06-22 15:27:16 +02:00

test_frontend_retention.py

fix(contributor-scan): retain declared Gruntfile and report census gaps honestly

2026-09-14 08:07:06 +02:00

test_fusion.py

feat(fusion): Phase 13 — two-panel fusion adjudication

2026-06-22 18:57:03 +02:00

test_gate.py

feat(hardening): Phase 15 — redaction, retry/backoff, budgets, threat-model docs

2026-06-22 20:02:34 +02:00

test_gitignore_semantics.py

fix(preprocess): read .gitignore with git's semantics; never ignore a tracked file

2026-09-25 15:29:43 +02:00

test_graph_reuse.py

feat(pre-push-safety-net): reuse compatible artifacts throughout replay

2026-09-13 17:08:49 +02:00

test_hardening_pipeline.py

feat(hardening): Phase 15 — redaction, retry/backoff, budgets, threat-model docs

2026-06-22 20:02:34 +02:00

test_harness_ext.py

feat(harnessx): configurable provider + per-task budget caps

2026-06-22 17:59:44 +02:00

test_harness.py

ci: add quality gates and release workflows

2026-06-18 10:06:21 +02:00

test_hunter_curated_tools.py

fix(hunter): ask for the final answer in plain text; DeepSeek's JSON mode lost it

2026-09-28 20:10:57 +02:00

test_hunter_harness.py

feat(harnessx): configurable provider + per-task budget caps

2026-06-22 17:59:44 +02:00

test_hunter_tools.py

perf(hunter): read a repository's source once per state, not once per tool call

2026-09-28 18:30:09 +02:00

test_hunter.py

feat(hunter): add standard-mode scheduling

2026-06-18 10:59:25 +02:00

test_improve.py

feat(pair-corpus-honesty): score non-vendored pointer pairs through a local cache

2026-09-05 10:45:43 +02:00

test_independent_population.py

spec(pre-push-safety-net): reserve the independent population v2 (17 cases, unscanned)

2026-09-26 17:53:51 +02:00

test_index.py

fix(ci): run ruff format, and record run 5 at 90% pair-correct

2026-09-08 19:46:58 +02:00

test_layout_facts.py

feat(pre-push-safety-net): exclude vendor inputs and partition frontend coverage

2026-09-13 11:53:29 +02:00

test_map_gate.py

feat(three-stage-scan): gate stage-2 maps against split-sink ranking

2026-09-04 11:53:06 +02:00

test_mapping.py

fix(mapping): a PHP method with a visibility modifier is a region

2026-09-09 17:33:36 +02:00

test_mcp.py

feat(mcp): the findings list carries the rung, and leads with what was decided

2026-09-09 13:15:54 +02:00

test_model_calibrate.py

fix(ci): run ruff format, and record run 5 at 90% pair-correct

2026-09-08 19:46:58 +02:00

test_model_change_context.py

fix(scan): symlink, unsupported-language and unparsed-file gaps are owned by what they name

2026-09-25 14:30:36 +02:00

test_model_config_value.py

fix(config): report every permissive setting in a file, not the first one

2026-09-20 16:30:04 +02:00

test_model_dominance.py

feat(model): an access-control finding names a line, and what the PDG does not give us

2026-09-09 13:09:47 +02:00

test_model_endpoint.py

spec(pre-push-safety-net): pre-register the hunter experiment on population v2

2026-09-28 09:10:02 +02:00

test_model_evidence.py

feat(model): shipped joins the evidence vector as a within-tier weight; v5 baseline

2026-09-11 13:45:00 +02:00

test_model_execution.py

feat(model-grounded-detection): group 6, the execution tier as a seam not a build

2026-09-08 16:58:03 +02:00

test_model_judge.py

fix(ci): run ruff format, and record run 5 at 90% pair-correct

2026-09-08 19:46:58 +02:00

test_model_ladder.py

fix(ci): run ruff format, and record run 5 at 90% pair-correct

2026-09-08 19:46:58 +02:00

test_model_partitions.py

fix(contributor-scan): retain declared Gruntfile and report census gaps honestly

2026-09-14 08:07:06 +02:00

test_model_pipeline.py

feat(model): a region that spans the flow, not just a function

2026-09-09 09:27:19 +02:00

test_model_regions.py

fix(ranker): scope a region to the file that defines its handler (M1b 12.4)

2026-09-17 14:04:39 +02:00

test_model_report.py

fix(ci): run ruff format, and record run 5 at 90% pair-correct

2026-09-08 19:46:58 +02:00

test_model_scan_stage.py

feat(contributor-scan): 1.3 the model layer reaches a user

2026-09-09 07:42:44 +02:00

test_model_scan.py

fix(contributor-scan): retain declared Gruntfile and report census gaps honestly

2026-09-14 08:07:06 +02:00

test_model_scope.py

fix(scan,taint): TypeScript's limit is owned by TypeScript; a source pattern is a token

2026-09-25 15:13:30 +02:00

test_model_shipped.py

feat(model): let the project say what it ships, and rank that first

2026-09-09 12:45:21 +02:00

test_model_specs.py

feat(repos): a WordPress plugin with a verified CVE, and the WordPress DB API

2026-09-09 16:13:17 +02:00

test_model_taint.py

feat(model): one finding per sink site, and WordPress's registration model

2026-09-09 14:30:48 +02:00

test_model_taxonomy.py

feat(reports): say what was looked for, and what cannot be decided

2026-09-09 15:08:01 +02:00

test_module_audit.py

feat(model-grounded-detection): group 1, subtract the noise architecture

2026-09-08 12:15:54 +02:00

test_mutation_fixtures.py

feat(three-stage-scan): add aliased-sink mutation the regex scan misses

2026-09-04 11:50:44 +02:00

test_obligation_check.py

feat(model-grounded-detection): group 1, subtract the noise architecture

2026-09-08 12:15:54 +02:00

test_obligation_dominance.py

feat(authorization-obligations): Dominance protocol with an order-based default

2026-09-05 19:38:04 +02:00

test_obligation_facts.py

feat(authorization-obligations): group sibling handlers and find consistency anomalies

2026-09-06 07:24:13 +02:00

test_obligation_intent.py

feat(authorization-obligations): hunter adjudicates intent without raising evidence

2026-09-06 07:24:13 +02:00

test_obligation_operations.py

feat(learning-harness): declared twins and a fixture-difficulty check

2026-09-06 13:09:56 +02:00

test_obligation_policy.py

feat(authorization-obligations): group sibling handlers and find consistency anomalies

2026-09-06 07:24:13 +02:00

test_obligation_reports.py

feat(authorization-obligations): reports and ranking name the obligation

2026-09-06 07:24:13 +02:00

test_obligation_scan.py

feat(authorization-obligations): run the checker in MAP with config and manifest counters

2026-09-06 07:24:13 +02:00

test_obligation_shapes.py

feat(authorization-obligations): obligation shapes derived from trusted pairs

2026-09-05 19:38:04 +02:00

test_obligation_siblings.py

feat(authorization-obligations): group sibling handlers and find consistency anomalies

2026-09-06 07:24:13 +02:00

test_opencode_skills.py

chore: add kiro spec-driven tooling and steering docs

2026-06-21 14:28:19 +02:00

test_pair_corpus.py

feat(model-grounded-detection): group 1, subtract the noise architecture

2026-09-08 12:15:54 +02:00

test_pair_harvest.py

feat(authorization-obligations): harvest mode handler_context and the obligation label

2026-09-06 07:24:13 +02:00

test_pair_hygiene.py

fix(ci): run ruff format, and record run 5 at 90% pair-correct

2026-09-08 19:46:58 +02:00

test_pair_registration_context.py

fix(ci): run ruff format, and record run 5 at 90% pair-correct

2026-09-08 19:46:58 +02:00

test_pipeline_calibration.py

feat(calibration): automatic in-pipeline false-positive feedback

2026-06-21 20:44:08 +02:00

test_policy.py

feat(governance): Phase 1 — CWE policy store + project score (zero-dep)

2026-06-22 12:31:35 +02:00

test_preprocess.py

feat(scan): add initial SAST pipeline

2026-06-17 21:13:51 +02:00

test_profiling.py

perf(push): drop per-entry cache fsync; let long runs profile themselves

2026-09-23 19:05:08 +02:00

test_provenance.py

feat(pair-corpus-honesty): deterministic provenance fingerprint at scan time

2026-09-05 10:45:43 +02:00

test_provider_retry.py

fix: a body that stops mid-read is transient, and a round that dies still says so

2026-09-07 07:37:41 +02:00

test_push_admission.py

fix(pre-push-safety-net): take the operation symbol from the call (M1b 12.1)

2026-09-17 09:56:59 +02:00

test_push_assistance.py

feat(pre-push-safety-net): bound optional assistance to admitted witnesses

2026-09-13 18:03:31 +02:00

test_push_cache_semantics.py

feat(pre-push-safety-net): key evidence reuse by compatible semantics

2026-09-13 16:16:19 +02:00

test_push_cache.py

fix(push): stop rescanning the whole cache on every publication

2026-09-23 14:12:47 +02:00

test_push_contracts.py

feat(pre-push-safety-net): retain bounded change and rename context

2026-09-12 21:12:45 +02:00

test_push_eligibility.py

feat(pre-push-safety-net): enforce evidence eligibility and record blocked qualification

2026-09-13 19:32:43 +02:00

test_push_experiment.py

feat(pre-push-safety-net): measure frozen PMPro and Node replay feasibility

2026-09-13 15:24:06 +02:00

test_push_feasibility.py

feat(pre-push-safety-net): measure frozen runtime feasibility and retain no-go

2026-09-13 19:06:02 +02:00

test_push_hook.py

feat(pre-push-safety-net): preserve existing hooks during explicit integration

2026-09-13 17:51:28 +02:00

test_push_inputs.py

feat(pre-push-safety-net): pin labeled evaluation inputs

2026-09-12 18:45:54 +02:00

test_push_interface.py

feat(pre-push-safety-net): execute full Git push transactions

2026-09-13 17:45:16 +02:00

test_push_policy.py

fix(push): a gap from another language partition no longer blocks admission

2026-09-21 13:50:38 +02:00

test_push_regression_integration.py

feat(pre-push-safety-net): cover combined push failure paths

2026-09-13 18:43:13 +02:00

test_push_report.py

feat(pre-push-safety-net): render bounded reports and preserve full evidence

2026-09-13 13:25:44 +02:00

test_push_reuse.py

feat(scan): size portions from measured per-answer cost and isolate the kill culprit

2026-09-23 07:55:33 +02:00

test_push_runner.py

feat(pre-push-safety-net): reuse compatible artifacts throughout replay

2026-09-13 17:08:49 +02:00

test_push_scoring.py

feat(pre-push-safety-net): preserve uncertainty in benchmark scoring

2026-09-12 19:04:20 +02:00

test_push_snapshot.py

feat(pre-push-safety-net): retain bounded change and rename context

2026-09-12 21:12:45 +02:00

test_push_vetoes.py

feat(pre-push-safety-net): render the experimental evaluation explanation (M1b 10.10)

2026-09-16 19:59:12 +02:00

test_rank.py

feat(harness): add runtime and reachability mapping

2026-06-18 07:57:06 +02:00

test_redaction.py

feat(hardening): Phase 15 — redaction, retry/backoff, budgets, threat-model docs

2026-06-22 20:02:34 +02:00

test_regress_candidates.py

feat(three-stage-scan): select bounded regress candidates with sev-5 extras

2026-09-03 22:36:02 +02:00

test_regress_recipes.py

feat(three-stage-scan): add language recipes with inconclusive fallback

2026-09-03 22:48:10 +02:00

test_regress_safety.py

feat(three-stage-scan): reject unsafe regression snippets before execution

2026-09-03 22:12:45 +02:00

test_regress_verdicts.py

feat(three-stage-scan): map sandbox outcomes to worth-fixing verdicts

2026-09-03 22:59:01 +02:00

test_reports.py

feat(cpg): give the files a frontend refuses a CPG of their own

2026-09-10 07:32:20 +02:00

test_repos.py

feat(repos): pinned known-vulnerable checkouts, offline by default

2026-09-09 08:10:59 +02:00

test_ruleset.py

feat(benchmark): Phase 2 (task 4) — actionable per-rule signals, ledger, FP knob

2026-06-22 15:41:17 +02:00

test_run.py

feat(scan): add initial SAST pipeline

2026-06-17 21:13:51 +02:00

test_sandbox_probe.py

feat(three-stage-scan): run or skip deep regression with visible degradation

2026-09-03 23:43:15 +02:00

test_sandbox_runner.py

feat(three-stage-scan): run or skip deep regression with visible degradation

2026-09-03 23:43:15 +02:00

test_scan_deep.py

feat(propose-adjudicate-prove): overlay adjudicates inventory before prove

2026-09-04 13:57:33 +02:00

test_scan_pipeline_integrity.py

feat(contributor-scan): 1.3 the model layer reaches a user

2026-09-09 07:42:44 +02:00

test_score_pipeline.py

feat(ruleset): Phase 2 (task 3) — rules as governed data, severity from policy

2026-06-22 15:27:16 +02:00

test_scoring.py

feat(benchmark): Phase 2 (task 4) — actionable per-rule signals, ledger, FP knob

2026-06-22 15:41:17 +02:00

test_semantic_cst.py

feat(tree-sitter-overlay-extra): parse JS/C/Java overlay via optional extra

2026-09-04 15:07:35 +02:00

test_semantic_extra.py

feat(tree-sitter-overlay-extra): parse JS/C/Java overlay via optional extra

2026-09-04 15:07:35 +02:00

test_semantic_overlay.py

feat(tree-sitter-overlay-extra): parse JS/C/Java overlay via optional extra

2026-09-04 15:07:35 +02:00

test_semantic_pipeline.py

fix(ci): run ruff format, and record run 5 at 90% pair-correct

2026-09-08 19:46:58 +02:00

test_skills.py

feat(skills): Phase 14 — Trail of Bits skill index + router

2026-06-22 19:51:04 +02:00

test_slot_contract.py

feat(slots): Phase 3 task 6.3 — model-disabled, slot-contracted deterministic stages

2026-06-22 18:39:09 +02:00

test_split_sink_fixtures.py

feat(three-stage-scan): add split-sink fixtures the regex scan misses

2026-09-04 11:49:44 +02:00

test_split_sink_regress.py

feat(three-stage-scan): prove split-sink triggerable in sandbox

2026-09-04 11:54:01 +02:00

test_stages.py

feat(three-stage-scan): keep quick as stage-1 inventory

2026-09-03 23:14:00 +02:00

test_subtraction_complete.py

fix(ci): run ruff format, and record run 5 at 90% pair-correct

2026-09-08 19:46:58 +02:00

test_tool_hunter.py

fix(hunter): ask for the final answer in plain text; DeepSeek's JSON mode lost it

2026-09-28 20:10:57 +02:00

test_typescript_grammar.py

feat(learning-harness): absence rows travel with their registration

2026-09-06 12:39:55 +02:00

test_verification.py

ci: add quality gates and release workflows

2026-06-18 10:06:21 +02:00

test_verify_judge.py

feat(harnessx): configurable provider + per-task budget caps

2026-06-22 17:59:44 +02:00