- a saved integer past 2^53, which JSON parsing would round to another id,
fails the step that saves it
- request paths and JSON strings read only `{saved:name}` as written; the
percent-encoded spelling is read only in a goto URL, whose parsed form
spells braces that way
- a held slug that neither `&` spelling gives counts as unknown: a create
whose slug spells `&` learns no slug rule beside it, and no later create
widens the spelling
- `save_response.path` takes the request path's portable pattern, shared as
START_ORIGIN_PATH, so JSON Schema clients compiling it with the `v` flag
read it
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Preserve XML, OpenStep and binary dictionary keys at their format owners. Share validated binary reference facts with materialization while retaining scalar precision, UID identity and bounded expansion. Migrate archive, Interface Builder, native and app-bundle consumers, then remove obsolete key-omission accounting and schema copies.
Validated source regressions and compiled CLI/MCP workflows, including native plist fidelity and archive identity. Typecheck, lint, docs and test discovery pass. No performance benchmarks were required.
Preserve observed failed-response HTTP status through Linux Hopper setup diagnostics without changing download rejection, retry, or installation policy. Independently validated on latest main through red/green regressions, actual TLS sockets, focused checks and CI. Closes#7182.
Keep binary_session available when Linux owned-child probes lack procps ps. Preserve typed ownership-prerequisite failures in direct execution and report per-provider readiness in status. Independently validated on latest main through real stdio MCP plus focused regression, static and documentation checks. Closes#7164.
Upgrade Zod to 4.6.6 and refresh locked license inventories. Preserve REA's established timestamp spellings through one shared UTC/offset grammar, including completion-verifier consumers. Size the stdin resource regression from the child's actual heap capacity.
Fixes#7173
Includes the dependency and inventory changes proposed in #7184.
Co-authored-by: tijo95 <georgesarmede@gmail.com>
Scope MCP discovery, setup, doctor, status and recovery guidance by selected tool families using one contract inventory and shared catalog identity. Preserve default discovery, canonical validation and complete Evidence.
Fixes#3246. Builds on #3026. Real libcrypto workflow verifies all 5,367 exports and identical complete Evidence. Static/source/build/docs/Windows/package/Greptile checks pass; ten assertions reproduced identically on clean main and feature across Node 24/22, with cancelled coverage explicitly unverified. Merge uses the user-authorized bypass for baseline failures and unchecked human-readiness declarations, without attesting human review.
Co-authored-by: 0xLLVEL <fakhrezan1@gmail.com>
Co-authored-by: Codebuff <noreply@codebuff.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Add opt-in bounded output-schema advertisements while preserving the full default, canonical server validation and complete structured content. Verify real stdio discovery under a 1 MiB receive cap, all 5,367 libcrypto export coordinates, complete Evidence parity, whole-task serialization estimates and resource bounds.
Fixes#2512. Integrates #7054 by Akshad Jogi with original authorship retained.
Co-authored-by: Akshad Jogi <akshadjogi@gmail.com>
- Basic credentials and values saved with `secret: true` are redacted however
short, as declared secrets are; page values keep the 8-character floor
- a literal Basic password other than empty is refused, since Evidence
records literals as written
- `save_response` takes only a request the step starts, so an earlier
autosave cannot answer it, matches its path as the browser spells it, and
stops watching requests when the step ends
- saving reads are bounded: a request step streams its answer up to
4,000,000 characters; a UI step refuses a larger received size unread
- goto placeholders fill the destination URL before structured query values
join it
- reading a composed string stops at the reading cap before comparing or
allocating more
- a slug's `&` spelling follows every create the plan accepted and every held
entity; direct slug rules and composed strings take it from held entities
when the create slugs none, and a slug rule they contradict is not learned
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The skill asks agents to delete what an earlier attempt created before
recording again, since the environment holds a leftover as seeded data
(EspoCRM's fresh run). The spec and verification record the write-model
fixes, the same recordings rebuilt before and after them, and a fresh
recording of every application.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Recording through the UI let Grafana generate folder uids, and three write
model gaps then served folders whose address kept the first folder's slug:
- a template around a generated value won over a composed string that also
reads the title's slug; the composed string now wins when it reads a field
besides the template's value
- later creates checked composed strings against their own answers, but not
templates, so coverage served a template a later folder contradicted; a
template is now checked the same way
- a slug dropped `&`, which Grafana's Go slugs spell as the word; a slug rule
or part learns `ampersand: "and"` from the call that shows it, and calls
that slugged no `&` agree with either spelling
An address whose uid equals its title's slug could come from either field.
The entities the type holds now pick the one reading they share, and when
they cannot, a later create that tells the fields apart lends its composed
rule to the first create, whose body contract is learned again with it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Outcome, plan, tasks and verification for saved values, save_response
and Basic credentials, with the agent-built comparison against main.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Allow native matching representation budgets to be selected at runtime or launch while retaining the 96 MiB default and main shared-provenance accounting. Preserve actionable configuration recovery through the setting owner and verify complete CLI/MCP matches for 20k/50k synthetic inventories and real stripped LLVM 14/15 inventories.
Fixes#6406.
Based on #7112 by Kaoru0822; original authorship and co-author credit retained.
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
- click and press steps take save_response {method, path, save}: the
recorder waits, before the step acts, for the 2xx JSON response to the
request it triggers and saves values from it
- {saved:name} sends a saved value in a request path (and its query), a goto
URL, and a request's JSON body, where a string that is only the
placeholder keeps the saved JSON type; header values and Basic
credentials parts take {source: "saved", name}
- Basic credentials take parts: [username, password], each a literal,
secret, saved, cookie or local storage value, so the schema repeats the
part union once; path_params and json_values are gone, which keeps the
compact MCP catalog under its 1 MiB frame
- validation refuses a saved value no earlier step saves; other braces are
sent as written
- Evidence records an empty save list as absent, so scenarios that save
nothing record the same parameters and scenario_sha256 as before
- environment-build warns about saved values named like credentials that
are not marked secret (credential-named-saved-values)
- the skill writes the way the UI does and falls back to request steps;
skill version 50
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This reverts commit 7336a566ac (#7119).
After #7120 and the other merges of 2026-10-25 grew inspect_analysis_view's
complete input schema (71,762 bytes plus 3,003 bytes of examples), the 64 KiB
fitting from #7119 could no longer keep instance level 3 and fell back to
level 2. That summarized the inline Evidence source and the view selector
collections, so main CI failed in toolInputReferences, toolSchemaValidity,
analysisViewMcp, absoluteLocalPathSchema, javascriptFactsViewMcp and
godotPckMcp. #7119's own CI ran against bc143125e, before that growth.
Reverting restores the complete full-profile advertisement while the size
limit is redesigned against these contracts (#4081 stays open).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Refs: #7135
Scope expected Ghidra input failures to the active request after the authenticated startup handshake. Preserve operation output, identity and log coordinates while retaining cumulative output with its session owner; lifecycle and unexpected failures keep full diagnostics.
Share remote input-code classification with domain error projection. Verify startup rejection, healthy continuation, cleanup, unknown errors, OOM diagnostics and real CLI/MCP workflows. The measured unknown-symbol reply falls from 13,170 to 2,209 bytes with unchanged successful facts and Evidence IDs.
Add optional native load-image summary delivery and retained selected-row pages through shared CLI/MCP analysis views. Preserve source verification, provenance, unavailable collections, exact row identity and opt-in byte/check payloads.
Preserve direct caller paths during shared CLI summary publication and cover raw/DOL projection contracts through the actual SDK. Real Ghidra ELF CLI/MCP captures, retained and file-backed pages, focused regressions and static checks passed; the PR records existing main schema failures and the JavaScript resource-recovery CI uncertainty.
* fix(mcp)!: return a compact binary_session status by default
binary_session {} with no target open returned about 66 KB, most of it
provider rejection diagnostics and candidate capability sets that the
skill's recovery flow re-reads on every retry. 6.8.0 grew it by 21%
because the shared regex capability limitation began repeating setting
bounds and step-by-step recovery for every search operation of every
provider (#6099, #6267, #6301, #7049) and IDA gained that limitation for
its two search operations (#4862).
The regex limitation now states the effective deadline and heap, their
sources and where to change them; the deadline and heap errors keep the
full recovery. binary_session gains detail: summary (default) or
complete. The summary keeps open state, target identity, the binding,
route capabilities, run, activity and health status, each candidate's
availability and target-support status, code and reason, and every
tool's availability, reason, remediation and modes, and lists the
fields it leaves out in omitted_fields. Complete detail returns the
previous shape. The CLI capabilities and providers commands keep
complete provider facts through the same application projection.
BREAKING CHANGE: binary_session {} no longer returns
analysis_provider_candidates[].availability.diagnostics,
target_support.diagnostics or capabilities, or tool_availability group
surface, effects and annotations. Pass {"detail": "complete"} to get
them.
Fixes#7109
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(mcp): key binary_session output fields by detail
The output schema accepted detail: "complete" with summary candidates or
tool groups, and a summary without omitted_fields, so detail did not
tell callers which fields existed. Each session kind is now crossed
with a detail-specific shape: a summary requires omitted_fields and
summary-only candidates and tool groups; complete requires diagnostics,
candidate capabilities and static tool flags and forbids omitted_fields.
Both the canonical schema and the SDK-validated advertised schema reject
mismatched shapes.
The verifier helper accepts summary candidates only for a validated MCP
summary; CLI provider output still requires candidate capabilities.
Regex limitation tests check values, sources and setting names instead
of whole sentences.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): report JSON Lines selection refusals as caller errors
A --format jsonl refusal because the selected value is not a record
collection is the caller's output selection, not unreadable analysis
output. Report it as invalid_request/invalid_input that names the
selection and its observed shape, keep the --format json /
--filter-output remediation, and drop the rea doctor advice.
AnalysisInputError gains an optional producer remediation, mirroring
the other refusal classes. The JSON Lines owner examines its documented
boundaries once for both admission and refusal facts.
docs/cli.md now states that columnar results such as
inspect-sqlite-database --table rows are not record collections.
Fixes#7110
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): carry JSON Lines refusal facts as structured observations
JSON Lines refusals now put the location and shape of each refused value
in details.issues[].observed, and the message is built from those facts.
Tests assert the classification, issue path, observed facts, and the
shared remediation constant instead of fixed sentences.
The filtered analysis-view export check now uses a non-empty request
and compares the emitted rows with the unfiltered export.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(mcp): fit every advertised input schema within the 64 KiB client limit
Claude Desktop drops a local connector tool whose compact-JSON input
schema exceeds 64 KiB. The full input profile advertised
compare_web_captures at 171,867 bytes and inspect_analysis_view at
71,632 bytes, because the self-contained projection inlines the
complete producer capture schemas once per side.
The full advertisement now fits the limit at its producing boundary.
A property whose schema repeats an earlier sibling's points at it, and
schemas below the deepest instance level that fits are summarized as
their types, size constraints and descriptions. Every rewrite widens the
advertisement through the existing relaxation rules for oneOf, not,
if/then/else, contains and unevaluated keywords, so canonical inputs
stay accepted and the canonical contract still validates every call.
Schemas within the limit are unchanged, and the compact profile is
still presented from the complete projection.
Numeric const and enum literals are also relaxed in the advertisement,
like boolean literals, since Gemini accepts only string enum members.
Fixes#4081
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(mcp): keep fitted input summaries widening and at the deepest fitting level
Array summaries took item types only from `items`, so a tuple such as
`{prefixItems: [{type: string}], items: {type: number}}` was advertised
as a number array that rejected `["a", 1]`. Item types now include tuple
positions from `prefixItems`, a draft-07 `items` array, and
`additionalItems`, and stay unconstrained when later elements are open.
A summary can also be longer than the schema it replaces (a nullable
array becomes an anyOf), so candidate size is not monotone in the
retained depth and stopping at the first overflow could skip a deeper
candidate that fits. A schema is now kept whole when its summary is not
smaller, and fitting searches from the deepest level down.
Property tests check that every summary accepts what its schema accepts
over every catalog input subschema and synthetic tuples, and that the
fitted schema is the deepest fitting candidate. The browser comparison
boundary test now expects a malformed `after` capture to be rejected
canonically, since `after` is advertised as the same schema as `before`;
the same malformation on `before` stays advertised-invalid.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(javascript): bind defaulted and destructured Webpack factory roles
A Webpack factory whose module parameter was defaulted (`e = {}`) or
destructured (`{exports: x}`) reported an exact empty export list with
complete coverage, because roles were taken only from plain identifiers.
Bind factory roles once, by position, in javascriptBundlerFactoryRoles.
Defaulted module/exports parameters keep their role (the convention always
supplies the argument), `{exports: alias}` binds the alias to
module.exports, and a repeated sloppy parameter name binds its last
argument. Other role patterns (rest, array, nested or computed keys, a
non-identifier loader) receive no role and count as an unknown finding,
so the file's analysis is partial instead of claiming no exports. The
export collector, the factory require name and the standalone bootstrap
externals all use these parsed roles.
Fixes#7113
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(javascript): keep factory receivers through unordered reads
A defaulted role parameter kept the unused default in the binding facts
passed to write ordering, so a receiver read inside a loop, nested block
or closure was unresolved and its export silently dropped while the
module stayed complete.
The role owner now removes defaults the calling convention never runs
before ordering. Separately, an export-shaped write through a role
binding whose read stays unordered now counts as an unknown finding
instead of being dropped, so an incomplete list is never reported as
complete.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(javascript): treat never-run factory parameter defaults as dead code
Removing unused defaults from binding facts let writes inside a default
expression reach the export collector: `(e = {}, t = (() => {
e.exports.ghost = 1; })())` reported `ghost` although Webpack supplies
both arguments. Loader calls in such defaults also produced require
references and async chunk keys.
The role owner now records the default of every role parameter, any
binding pattern included, as never run. The export collector, the
async-chunk walk and the static source traversal skip those subtrees,
so they yield no exports, references or chunk keys. Defaults of later,
non-role parameters still run and are still reported.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* perf(native): charge matching for what it copies, not borrowed retained inventories
Native function matching deep-walked every retained list_procedures record
at about 2.8 KB per Ghidra-shaped row, although the session ledger lends
those immutable records without a copy, and charged each inventory's one
Evidence ID once per function. The report and Evidence copies of every
emitted member were not charged at all.
Charge each inventory row before parsing for the member it becomes,
including its report and Evidence copies, charge the shared inventory ID
once, and skip the walk only for borrowed immutable inventories and closure
records that matching never parses. Inline Evidence, bundle records and
selected observations keep their existing charges. Retained references now
fit about 22,000 rows per side instead of 10,000, and an oversized side is
refused before its rows are parsed.
Refs #6406
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(native): charge the envelope of borrowed matching inputs
Borrowed inventories skipped admission entirely, but matching copies their
limitations into the report and Evidence and digests their profile, so a
retained inventory with few rows and large limitations escaped the
allowance. Charge every envelope field and leave only the result payloads,
which row charging covers or matching never parses, uncharged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Reduced compact input schemas gave every array property an any-value item
schema, a six-member type list beside nested `items`. Gemini function
declarations accept `items` only beside the single `array` type, so every
request through a Gemini-backed client failed with
`properties[functions].items.items: field predicate failed: $type == Type.ARRAY`.
Reduced array properties now advertise the element types their canonical item
schemas accept. Unknown, tuple, and mixed array elements use the empty item
schema, and properties that may be arrays or another type list typed anyOf
branches. An inline `not: {}` keeps the empty schema. The compact catalog test
now rejects `items` beside a non-array type as well as untyped item unions.
Fixes#5201
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
inspect_macho reported only the symbol-table spelling (`_main`), while
LLDB, dlsym and observe_native_calls use the runtime spelling (`main`).
Passing a static name as a breakpoint silently selected a different
function, so results from the two tools could not be joined by name.
Keep `name` as the exact symbol-table fact and add a derived
`runtime_name` to every export and import row, produced by one shared
Mach-O symbol-name owner. Managed P/Invoke verification now uses the
same owner instead of its own underscore stripping.
Fixes#5298
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Retain complete browser scenario Evidence while delivering compact summaries and selected step, event, transaction and text views through shared CLI/MCP workflows. Preserve initial CLI failures across cleanup retries and validate with real Chromium/TodoMVC.
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(environment): take created entities' read-backs and re-derived update members
- A create's observation takes the fields the first read of its entity
shows that the answer lacks, and an item read's clock spellings, when the
read agrees with the answer otherwise. The plan records the read
(observation.read with its shape); an item read then observes rather than
checks, a list read only adds fields and still checks. Serving merges the
same recorded objects.
- An update member sent back as the clock the entity held and stored as a
fresh clock near the call is a clock rule; one stored as the slug of
another member stored as sent is a slug rule. Both are accepted with any
value, as whole-object writes send what they read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): record the write-views verification
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): reconcile write views without hiding contradictions
* fix(environment): reject incompatible slug-only saves
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(ghidra): report a DOL language override conflict as a configuration error
A REA_GHIDRA_LANGUAGE_ID outside PowerPC:BE:32:* made an otherwise admitted GameCube DOL an unsupported target. docs/ghidra-provider.md and the other fixed-import lanes treat such an override as a configuration error naming the conflicting setting. The DOL lane now uses the same languageOverrideConflict owner and names only REA_GHIDRA_LANGUAGE_ID, since DOL accepts a compiler-spec override.
* test(ghidra): keep a valid compiler spec out of the DOL language conflict
* feat(binary-layout): admit ELF32 i386, ARM and MIPS layout profiles
inspect_binary_layout refused every ELF32 object, so router and IoT
firmware (ARM32, MIPS32) got no layout facts at all.
The pwntools adapter now admits EXEC/DYN/REL objects in these profiles:
ELF64 x86-64, ELF32 i386, ELF32 little-endian ARM and ELF32 little- or
big-endian MIPS. Header fields are read in the declared byte order, the
ELF32 header is bounds-checked at 52 bytes, and pwntools' own reported
architecture, class and byte order must match the admitted profile.
Other machines, classes and byte orders keep an unsupported_target
refusal that names the admitted profiles.
The output schema admits exactly those architecture combinations, and
the Evidence subject now names the reported machine instead of always
claiming x86_64.
verify:binary:layout cross-compiles i386, ARM, MIPS and MIPSel
executables and relocatable objects (never executed) and checks every
section name, address, offset and size against GNU readelf through CLI
and MCP. The real-binary-layout workflow installs those toolchains.
Refs #6934
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* ci(binary-layout): install target libc headers for ELF32 fixtures
--no-install-recommends skips the libc6-dev-*-cross packages the cross
compilers only recommend, so the i686 fixture fell back to the host's
/usr/include and failed to find bits/libc-header-start.h. Install the
four target libc development packages explicitly.
Refs #6934
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(binary-layout): size the minimum symbol entry by ELF class
The relocation reference check required a 24-byte Elf64_Sym entry, so
every positive symbol reference in an ELF32 report was refused as
unreadable output. Elf32_Sym is 16 bytes; take the minimum from the
reported class. A 64-bit table with 16-byte entries is still refused.
Refs #6934
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(binary-layout): compare ELF32 symbols and relocations with readelf
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Apple's crash report JSON documents termination.byPid, byProc and flags, the body's parentPid, and
the metadata line's build_version and platform. The parser shapes did not list them, so they were
reported only by name in unlisted_keys / unlisted_body_keys and their values were dropped: a
watchdog kill no longer said which process ended it, and process kept parentProc without its pid.
#4297 asks for termination members beyond namespace/code/indicator to be retained as recorded, and
docs/apple-ips.md says documented fields are reproduced.
They are now kept: termination.by_proc, termination.by_pid and termination.flags, process.parent_pid
and metadata.build_version / metadata.platform. The integers are recorded integers, so they count
toward max_integer_text_bytes and a type mismatch is recorded like the other fields.
* fix(cli): preserve literal Markdown table facts
* test(cli): retain capacity diagnostics through Markdown escaping
* fix(cli): keep intraword underscores readable in Markdown tables
Underscore runs between letters or digits cannot open or close CommonMark emphasis, so escaping them only adds noise to identifiers such as source_index and invalid_request in raw Markdown. Escape underscore runs only at word boundaries and restore the original diagnostic assertions.
