Commit Graph
3422 Commits
Author SHA1 Message Date
01e5c59551 fix(experiment): validate native engine pins before model trials
Preserve the rejected cross-function launch caused by a non-embedded Cargo test build replacing the mutable CLI. Check explicit engine identity before profile creation and after the task; use an independent executable copy for measured runs.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 08:03:30 +00:00
33c4282b8a feat(experiment): expose structured native call and argument relations
Keep recovered callee identities separate from fspec annotations and SSA argument values separate from passing storage. Add selected calls and function-input reads through the shared fact owner, with real SQLite MCP coverage and a pre-pinned cross-function model task.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 07:58:01 +00:00
f34425dc24 fix(experiment): enforce Code Mode payload budgets on guest errors
Reject oversized error delivery with explicit omission metadata, account for JSON escaping, and retain source observations and their lookup after failure. Reproduce and verify the real MCP path without claiming model-token savings from byte counts.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 07:33:09 +00:00
dad0228df8 experiment(re): verify model composition of class and dispatch tables
Record matching six-byte table oracles, real byte-tool use, model response counts and whole-task costs without generalizing a single paired observation.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 07:23:40 +00:00
73db4fad54 feat(experiment): compose bounded native byte windows with RE facts
Add shared SDK/CLI/MCP byte reads over the existing Kuna data protocol, preserve partial batches and exact u64 ranges, and retain complete source Evidence. Reuse named read lifecycles, bound aggregate hex expansion, and verify stock/native/direct MCP paths plus a shipped-file byte oracle before pinning a table-dispatch model task.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 07:17:18 +00:00
c53c2a9b90 experiment(re): record equal-coverage model use of retained Pragma code
Keep actual response/MCP counts, whole-task token usage and resource receipts separate. Both adapters complete the local task with explicit missing IR; no general speedup is inferred.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 06:58:46 +00:00
9e1b39358a experiment(re): pin a model task over recovered Pragma code
Exercise the restored code facet with a question fixed before trial and independent instruction anchors, using identical envelope guidance in both atomic calling modes.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 06:52:03 +00:00
1346e97dca fix(experiment): preserve code when encoded native graphs exceed capacity
Bound compact native JSON with the existing writer, keep a valid code facet when the graph cannot fit, and pass the supervised byte budget explicitly. Verify real SQLite Parser/Pragma captures, exact preserved C/mappings/graph facts and public MCP reads. Keep string inventory parsing/projection under its own owner.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 06:50:20 +00:00
dc022bcd0f experiment(re): separate model rounds from MCP requests in fair trials
Validate per-response usage totals, retain the negative large-VM comparison and qualify the client executor, cache and investigation-depth differences. Add optional common envelope/stopping guidance without relabeling historical runs.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 06:29:26 +00:00
5aa78c873e experiment(re): pin coverage criteria for a large SQLite VM task
Record independent machine-code anchors before model trials, separate cache regimes and incomplete answers, and exercise real recovered-IR capacity failure without hiding retained code.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 06:09:46 +00:00
15d75ea1ea fix(experiment): preserve cached fact ownership for SDK callers
Freeze parsed inventories once at their owning boundary and share the rule with native IR and function views. Verify the real SDK mutation regression and retain corrected equal-depth SQLite model costs, including persistent-cache limitations.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 06:07:46 +00:00
e9b9a8dd0e experiment(re): describe direct atomic trials without a Code Mode tool
Keep the direct adapter benchmark instructions consistent with its actual registered tools; retain historical trial profiles unchanged.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 05:48:57 +00:00
f3f209d03f experiment(re)!: preserve partial captures and resolve function entries
Keep validated code available when native graph extraction is unavailable, bind output to the actual input/entry, preserve cancellation observations, and share batch source/error semantics. Add filtered function/alias reads and slice field selection across adapters.

Verified 38 focused tests, targeted lint, real stock/native MCP workflows, large SQLite graph-capacity fallback, and real cold/warm MCP reuse. Retain actual model costs and unequal-depth/selection limitations instead of claiming general superiority.

BREAKING CHANGE: Experimental native snapshots may succeed with readable code and an unavailable recovered_ir facet. Inspect facets before graph reads; unavailable graph counts are null/unknown.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 05:48:03 +00:00
4f13bd6733 experiment(re)!: separate IR origins from machine instruction facts
A real model conflated an optimized CFG edge with the bytes at its instruction origin. Make provenance explicit and add bounded, batched instruction windows from the existing Kuna CLI, preserving exact bytes, per-window failures and Evidence. Record the task-cost comparison without treating the incomplete machine-path answer as an equal-quality win.

Verified 33 focused tests, targeted lint, real stock/patched MCP decoding with mixed success, and a real 4096-row capacity run under the resource observer.

BREAKING CHANGE: Experimental operation.address/addresses and block.start become origin_address/origin_addresses and origin_start. Printed CFG target_address becomes target_origin; historical source captures are unchanged.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 04:51:55 +00:00
f2f029c5dd experiment(re): retain native model trace and task-cost limits
The real model uses recovered controllers and indexed phi/value relations, but the first native-IR trial still does not beat the traditional local-question workflow. Record actual whole-task usage and prepare value-origin and small-function counterexamples.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 04:33:55 +00:00
a7e9f8843b experiment(re): expose native CFG and value identities for composition
[AUTOMATED] Preserve the opt-in pinned Kuna exporter as a reproducible source patch. Share typed graph/value reads and derived control dependence through the SDK, CLI and MCP adapters, retaining unknown runtime and memory semantics.

Validated 29 focused tests, lint, actual diagnostic/native MCP workflows and a real zlib CLI composition. C and line mappings match the prior stock capture. Full upstream corpus/workspace gates and revised real-model costs remain unverified.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 04:28:22 +00:00
fcfc87b2b4 experiment(re): record actual three-way model task costs
Retain failed attempts, cached input accounting and payload-exit process-family counters. The first comparison favors the traditional workflow; keep the negative finding explicit.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 04:00:19 +00:00
c5cefd622b experiment(re): refine composition contracts from a real model trace
Record genuine Code Mode use and full model usage rather than response-only estimates. Describe return shapes, permit byte-budgeted internal C reads, separate guest CPU from analysis deadlines, and move subsequent model profiles outside the source tree. Verify lease, admission, cleanup-retry and cancellation semantics.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 03:28:30 +00:00
5aa1c5c64f experiment(re): expose shared atomic facts through isolated Code Mode
Add a compact rea-codemode MCP server, bounded QuickJS execution and shared SDK/direct adapters. Reuse native target and process owners; retain evidence and lease-scoped snapshots. Extend the real-model trial runner with an isolated experimental MCP registration. Capability and evaluation work remains in progress.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 03:17:05 +00:00
2e11fc23fd experiment(re): validate guard slices and record token tradeoffs
Verify three repaired zlib mappings, a SQLite call guard and all ordinary 16-bit header values against retained native evidence. Record producer omissions, failed-run costs and the small-function context counterexample. Preserve existing capture outputs with exclusive writes.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 02:59:46 +00:00
fd79a273f3 experiment(re): compose native SSA uses and branch dependencies
Retain bounded native snapshots and public MCP evidence, then query explicit SSA relations without interpreting reconstructed C. Preserve unknown flow, cloned block identity and dependency frontiers.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 02:59:46 +00:00
d711c55aed experiment(re): distinguish point facts from object selectors
Apply the interior-reference counterexample to the facade: label exact-address scope and emit location endpoints while preserving caller selectors separately. Carry canonical target-content provenance through experimental MCP replies and verify it on the real client. Keep original measurements pinned to the v1 response shape. Nine experiment checks and real Kuna lifecycle/context checks pass.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 02:59:46 +00:00
72b24433ea experiment(re): record native sharing and composition counterexamples
Measure twelve equivalent MCP navigation workflows, native reference plans on SQLite/zlib, a compiled interior-reference oracle and a failed exact code-mapping composition. Preserve failed attempts and measurement limits, including unavailable live-agent effectiveness evidence. Add bounded native probes and deterministic mapping tests. All 191 Kuna cases pass serially; nine experiment tests and one-checker typecheck pass.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 02:59:46 +00:00
2e19ece025 experiment(kuna): preserve typed reference facts at their owner
Expose source-owned reference kind, endpoints, function identities and instruction text for experiments. Keep the existing address-only public result and retained-cache shape; materialize richer facts only when explicitly requested. Record prior art, composition rules and the distinction between fact, transport and compute units. Verified 18 producer boundary cases and an independently checked native interior-reference example.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 02:59:46 +00:00
27f981352d experiment(re): compare fact granularity through real MCP
Add a non-shipping MCP facade over the existing admitted analysis port, real-provider comparison runner and offline token accounting. Keep the full catalog, source Evidence and per-input failures. Separate verification runs from equivalent-work timings. Kuna zlib comparisons pass; broader comparisons remain in progress.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 02:59:45 +00:00
0d99e8c5c5 experiment(re): define composable facts and context boundaries
Start a non-integrating experiment around fact identity, per-input outcomes, paging and shared primitive composition. Include deterministic boundary checks; real-provider comparisons remain in progress.

Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
2026-10-16 02:59:45 +00:00
Ysopwnable 42fc6c53ae fix(javascript): preserve incoming callback target uncertainty (#4243)
Preserve incoming callback uncertainty and resolve proven preceding writes.
Share read-position and initializer identity facts with returned-reference
analysis; cache completed projected work with actual cycle dependencies.

Fixes #3880
2026-10-16 10:09:22 +08:00
morluto 1bd5c5c87e fix(adb): retain and resolve owned hash cleanup
Keep exact returned hash and directory failures linked to their cleanup owners. Retain input staging while its descriptor remains unresolved, and resolve the linked errors only after successful release. Preserve typed cleanup and partial acquisition facts through file, package, and bugreport consumers instead of wrapping or flattening them away.

Validated 189 affected tests across 13 files, build, typecheck, lint, and docs. Three larger integration cases were excluded to respect the CPU limit. New real-file fault controls reproduce on the old source and verify persistent failure, owner identity, cancellation, retained output, and recovery. No physical ADB device or real JADX engine verification. Refs #4266, #4274.
2026-10-16 09:06:26 +08:00
morluto 6e74704f93 fix(android): resolve hash cleanup through its owner
Bind the typed hash failure to its retained descriptor and resolve cleanup uncertainty only when that exact owner closes. Preserve the original failure through immediate retirement and later shutdown recovery. Controlled real-file regressions cover immediate and persistent close faults; no real JADX engine was launched. Refs #4266, #4274.
2026-10-16 09:06:19 +08:00
morluto 3ed925e547 fix(errors): exclude recovered cleanup resources
Compose only unresolved primary resources into a later cleanup failure, retaining both original causes and diagnostics. Document exact-owner recovery and cover resolved and unresolved composition controls. Refs #4266, #4274.
2026-10-16 09:06:13 +08:00
morluto ca7ee018b2 fix(providers): own shared startup and retained cleanup
Give concurrent Ghidra and Hopper callers independent waits over one owned startup. Join final-caller cleanup, preserve failed cleanup owners, and derive readiness from the current startup result and live transport. Distinguish provider closure from caller cancellation and clean up interrupted final lineage observation.

Reuse the Ghidra socket fixture, replace stale cancellation health assertions with cold and warm reuse controls, and cover public MCP peer cancellation. Validation: 315 affected tests, build, typecheck, lint, docs, and 25 relocated startup controls passed. Controlled child bridges only; real Ghidra JVM and Hopper GUI verification remains unperformed. Refs #3983, #4227, #4274.
2026-10-16 08:51:11 +08:00
morlutoandVinícius DS Andrade 808339f0b1 fix(native): distinguish undecoded Swift spelling adjustments
Recognize known undecoded producer spellings without treating them as successful demangles. Preserve actual producer output and derive symbol types from the domain contract instead of duplicating its validation schema.

Compare retained output with the real demangler through MCP, while preserving CLI malformed-input and line-ending coverage.

Refs: #4215, #4287

Co-authored-by: Vinícius DS Andrade <[email protected]>
2026-10-16 08:26:19 +08:00
morlutoandHenglie 2f6b9ecdc6 fix(native): preserve slice identity when selecting UUID rows
Carry the selected slice name and the existing tool owner's dwarfdump spelling separately. Compare observed row labels exactly, preserving explicit ABI versions and uncertainty for missing rows.

Correct the production-boundary fixture and retain wrong-version and mixed-row controls. Real Apple multi-version selection behavior remains unverified.

Refs: #4195, #4288

Co-authored-by: Henglie <[email protected]>
2026-10-16 08:26:18 +08:00
morluto 5f0d18a534 fix(ghidra): retain request limits and startup cleanup
Preserve the full request options through the FIFO and settle interrupted or rejected final lineage observation through owned startup cleanup. Keep the startup diagnostic checkpoint until finalization succeeds.

Validate typed expired ping, cancellation, child exit, runtime removal, retry, and MCP recovery without running a native Ghidra instance.

Refs: #4289, #4274
2026-10-16 08:26:18 +08:00
morlutoandHemang Varshney 018ab2c5d9 fix(javascript): retain unresolved network targets with provenance
Resolve stable lexical and URL targets using shared effects, and preserve unresolved request expressions as partial facts. Carry their uncertainty through workers, artifact graphs, and the CLI. Scope partial endpoint identity to the source site.

Refs: #3451, #3534

Co-authored-by: Hemang Varshney <[email protected]>
2026-10-16 08:14:52 +08:00
morlutoandGeronimo 1b9b725c4c fix(javascript): unify loader binding and mutation ownership
Model the CommonJS wrapper binding and query shared lexical identity, mutation, and evaluation timing at each call. Remove duplicate importer interpretation and preserve uncertainty for with scopes, prototypes, repeated writes, and opaque object escapes.

Validate generated fixtures against bounded Node oracles; no analyzed target execution is added.

Refs: #3982, #4072

Co-authored-by: Geronimo <[email protected]>
2026-10-16 08:14:52 +08:00
morlutoandGeekwolf b67baede2e fix(electron): unwrap transparent inline window options
Use the shared expression owner for BrowserWindow options and webPreferences. Retain overwrite and spread uncertainty through CLI and MCP evidence.

Refs: #4250, #4281

Co-authored-by: Geekwolf <[email protected]>
2026-10-16 08:14:51 +08:00
morlutoandGeekwolf 38f92d108e fix(browser): match exact WebMCP receiver segments
Inspect AST property segments instead of flattened suffixes. Reject dotted-key and suffix collisions while retaining page-declared trust and source evidence through the CLI.

Refs: #4205, #4279

Co-authored-by: Geekwolf <[email protected]>
2026-10-16 08:14:51 +08:00
b3389a5b08 fix(mcp): preserve nested JSON values in compact array schemas
Give array unions an unrestricted item schema and validate converted tools/list schemas for both advertised profiles. Keep client constraints separate from JSON Schema validity.

Refs: #3838, #3921, #3842

Co-authored-by: Kaoru0822 <[email protected]>
Co-authored-by: 冰水泡枸杞 <[email protected]>
2026-10-16 08:14:50 +08:00
morlutoandalfredo1491 4fd28fc9e0 fix(setup): preserve JSONC trivia when removing registrations
Remove properties and their separators using parsed JSONC token spans. Preserve unrelated comments, BOM, CRLF, backups, and repeated removal behavior.

Refs: #4244, #4284

Co-authored-by: alfredo1491 <[email protected]>
2026-10-16 08:14:50 +08:00
morlutoandSyed Rishad d52a488476 fix(ghidra): retain scoped failures through process finalization
Drain output within caller bounds at asynchronous completion owners and classify JVM exhaustion from the failing operation only. Preserve raw-byte scope, credential redaction, configured heap provenance, cancellation and exact cleanup failures. Validation uses controlled processes and provider fixtures; native Ghidra OOM execution was not repeated.

Refs #4241, #4275

Co-authored-by: Syed Rishad <[email protected]>
2026-10-16 07:25:23 +08:00
morlutoandLiu Jie 43bc6e1d23 fix(javascript): preserve canonical rest-array property keys
Reuse the semantic array-index parser so leading-zero keys stay unresolved instead of selecting positional arguments. Preserve canonical indices, literal fallback and dynamic-key uncertainty.

Refs #4242, #4272

Co-authored-by: Liu Jie <[email protected]>
2026-10-16 07:25:23 +08:00
morlutoandogutdgn 89dd71f0f0 fix(analysis-view): preserve native relationship coverage
Carry the parent native coverage facts through inline and retained views independently of row-page exhaustion. Preserve uncertainty, optional diagnostics and source Evidence.

Refs #3891, #4263

Co-authored-by: ogutdgn <[email protected]>
2026-10-16 07:25:22 +08:00
morlutoandY0UW1SH a05092726f fix(go): distinguish dependency coverage from parse completeness
Explain unknown dependency coverage when embedded metadata has no main-module record. Preserve decoded dependencies and distinguish GOPATH-style omissions from observed module-mode empty lists.

Refs #4191, #4267

Co-authored-by: Y0UW1SH <[email protected]>
2026-10-16 07:25:22 +08:00
morlutoandPolly Labs 79e83cbd67 fix(setup): resolve empty Claude profile paths consistently
Use the default home registration and personal skill roots for an empty profile override. Share marker and skill path resolution while preserving nonempty caller paths, backups and idempotence.

Refs #4262, #4270

Co-authored-by: Polly Labs <[email protected]>
2026-10-16 07:25:21 +08:00
morlutoandLiu Jie 862a0b2631 fix(artifacts): reject colliding extraction paths before writing
Preserve duplicate archive occurrences in inventory and reject an ambiguous materialization before acquiring readers or output resources. Keep path collisions distinct from integrity failures.

Refs #4271, #4273

Co-authored-by: Liu Jie <[email protected]>
2026-10-16 07:25:21 +08:00
morluto 839410b663 fix(session): report retained file and candidate cleanup
Report cleanup from the owning snapshot and provider-candidate resource scopes without retrying or starting work. Keep status accurate before a client exists and through cleanup retry.

Refs #4277, #4274
2026-10-16 07:25:20 +08:00
morluto 93c620c3e7 fix(browser): own transport cleanup through finalization
Own CDP transports from admission through observed close, retain failed
owners for retry, and keep active peers separate from pending cleanup.
Preserve primary errors and completed facts through final cancellation.
Forward passive Electron CLI cancellation and dispose browser, runtime and
scenario owners from CLI and MCP shutdown.

Refs #4255, #4256, #4257
2026-10-16 06:46:26 +08:00
morluto 3b698645a0 fix(evidence): retain completed observations on late cancellation
Validate completed reports before cancellation projection, retain valid
Evidence and captured command diagnostics, and convert raw cleanup partials
at their application owner. Record partial Evidence consistently across MCP
handlers. Keep page inspection separate from unfinished bundle analysis.

Refs #4261
2026-10-16 06:46:25 +08:00
morluto 131fd5ddcd fix(filesystem): retain stable hash cleanup owners across callers
Preserve completed digests and primary failures when descriptor close fails.
Carry the exact handle owner through provider, readiness, cache, doctor and
verifier boundaries, and retry cleanup without rehashing or retrying a raw fd.
Close candidate-owned resources even when no target client was opened.
Keep verifier workspaces when provider shutdown remains incomplete.

Refs #4266
2026-10-16 06:46:25 +08:00