Attach to external browsers without implicit default-context overrides, and own temporary focus only on the selected page. Preserve unrelated page media and exact page inventory through CLI/MCP. Remove bootstrap-page retries; retain late/failed cleanup ownership and verified SDK notices.
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Parse shared request and response fields at the CDP boundary while preserving capture, session and runtime admission policies. Treat empty request IDs as unknown so unrelated events cannot be joined, and retain redirect occurrences and partial response evidence.
Validate 221 focused tests, fast checks, and real Chrome network and runtime verifiers.
Centralize CDP script/context declarations and frame-claim interpretation across web capture, Electron, V8, active runtime and storage seeding. Preserve each workflow's admission, provenance and retention policies while removing superseded helpers and schemas.
Malformed or conflicting frame claims now leave passive scripts unattributed. Reused context IDs replace future frame authority without rebinding earlier scripts, and source readers reject empty script IDs.
Validation: 153 focused tests and check:fast passed; 13 regression failures reproduced on the prior commit. Real Chrome script export and runtime CLI/MCP verifiers passed, including correctly reported uncertain cleanup after intentional target closure. Electron regression coverage used simulated CDP.
Parse each frame event using its declared protocol fields and retain loader and ancestry only across same-document updates. This removes alias-based attribution, invalidates malformed committed frame facts, and prevents stale WebMCP tools surviving a same-URL document replacement.
Migrate Electron, screenshot, capture, timeline, and execution consumers to the shared facts owner and retire the old helper exports. Validated 428 focused tests, test discovery, typecheck/lint, and real Chrome CLI/MCP network capture; Electron and WebMCP regressions use simulated CDP.
Retain file identity after Rust, web artifact, shared selected-input and historical capture reads complete before close or cancellation fails. Preserve descriptor-set snapshots in diagnostics alongside sanitized capture observations.
Classify canonical physical capture failures before cleanup composition; source permission errors and bare syscall failures keep their meaning through late cancellation and retry. Keep explicit sensitive-value redaction on retained facts.
Validation: 247 focused cases passed, two Linux-only cases skipped; build, typecheck, lint and formatting passed. Real descriptors with controlled finalization faults establish reader ownership and facts. Worker reply and multi-file database completion semantics remain separate audit work.
RFC 9110 section 5.5 excludes leading and trailing whitespace from a field value. The gRPC-web trailer parser stripped only leading SP/HTAB, so 'grpc-status: 0 ' was reported as '0 '. Trailing SP/HTAB is now excluded too; internal whitespace and the examined byte count are unchanged.
* feat: add FreeBuff agent support for REA setup
Register FreeBuff as a supported setup client using the shared user
agents directory (~/.agents/mcp.json with mcpServers, skills in
~/.agents/skills), which matches what FreeBuff Desktop loads. Adds the
client definition, boundary tests, supported-agent docs, and the skill
reference client list.
* fix(freebuff): detect FreeBuff from its profile directory, not shared .agents
The ~/.agents marker made FreeBuff appear installed for any home where
REA had installed shared skills (the default skill destination, also
Gemini CLI's), producing false doctor failures. Detect FreeBuff from
its own fixed profile directory ~/.config/freebuff-desktop instead,
and add a regression test covering a home with only the shared agents
skill directory.
* fix(setup): require product evidence for FreeBuff discovery
Address review findings at their producing boundaries and preserve directly affected consumers. Source inspection and git diff --check completed. Tests and automated checks were not run at the user’s explicit request.
* fix(uninstall): retain unverified shared client configuration
* fix(clients): preserve scoped removal evidence
* fix(doctor): omit undefined selectedClientIds under exactOptionalPropertyTypes
Passing selectedClientIds explicitly as undefined fails typecheck on this
branch (TS2379), which blocks the focused test workflow before any test
can run. Spread the option only when defined.
* fix(update): separate ownership evidence from product applicability
rea update's maintenance scope inspected only product applicability
evidence, so when a FreeBuff profile was absent but REA already owned a
version-pinned row in the shared ~/.agents/mcp.json, the plan omitted
freebuff and left the integration stale. FreeBuff now declares its
registration storage as shared, which proves ownership of an owned row
but never product applicability, and maintenance requests ownership
evidence through that declaration instead of reinterpreting the shared
paths. Doctor and setup keep readiness evidence, foreign rows stay out
of maintenance, and the Doctor exactOptionalPropertyTypes fix makes the
focused test workflow runnable.
* test(dotnet): split oversized operand extent diagnostics describe
The describe callback reached 162 lines (max 150), failing
max-lines-per-function and blocking the pre-push check for this branch.
Move the single-case extent scenarios into their own describe; no test
content or order changes.
* fix(update): preserve ZCode's disabled shared registration during maintenance
The shared user agents file doubles as ZCode's fallback, where
`enable: false` disables the registration. Ownership-based maintenance
read that row as a FreeBuff registration, whose format does not
recognize the flag, so the row was selected and the emitted setup
command rewrote it, dropping the flag and silently re-enabling a
registration the user had disabled.
registrationActivation now honors `enable: false` for any client whose
registration storage is shared, because the activation fact binds every
reader of that storage. Maintenance leaves such rows out; doctor
reports them as stale, matching ZCode's own treatment.
* fix(doctor): let explicit client selection authorize the config read
When an ACL or an inaccessible shared parent made every access probe
fail, a client selected by name was omitted before the configuration
read could return its unreadable-config detail, so doctor reported it
as not detected with installation remediation: an unresolved selected
path collapsed to absence.
An explicit selection now authorizes the read outright and the probes
serve automatic discovery only. A selected client with an unreadable
config reports invalid with the read failure; one with no config
reports missing; automatic discovery still probes before inspecting.
* fix(uninstall): share shared-storage checks between ZCode and FreeBuff
clients() queued ~/.agents/mcp.json twice: once as zcode_fallback and once
as freebuff. ZCode has no discoveryPaths, so its strict read still blocked
uninstall on malformed shared text, despite FreeBuff's new retain-and-continue
behavior. For a valid owned entry, ZCode removed it first, so FreeBuff reported
skipped rather than removed.
Make the shared file's checks and removal one shared decision:
- Add zcodeDiscoveryPaths to ZCode so it has its own applicability evidence.
- Gate zcode_fallback push on applicable, not just zcodeFallbackPath presence.
- Report shared storage ownership as one decision for both clients:
FreeBuff and ZCode now use clientOwnershipEvidencePaths together, so a
recognizable REA-owned row proves ownership once for the shared storage, and
both clients report the same result consistently.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(uninstall): merge duplicated host and prove shared ownership from owned rows
Uninstall.ts declared systemUninstallHost twice, preventing module load,
with the exported copy returning only clients and the second copy missing
clientWriteBlocker. Merge into one exported host carrying both blockers.
hasOwnershipEvidence used access() existence, so an unrelated malformed
~/.agents/mcp.json established ownership and blocked every removal.
Ownership now requires a readable, recognizable REA-owned row via the
injected filesystem, so unrelated shared data stays retained.
Also cover ZCode discoveryPaths in its path test.
* fix(uninstall): keep denied product-local configurations fail-closed
Unproven product applicability marked every configuration path of that client unverified, so a denied ZCode profile turned an unreadable product-local config into a retained item: uninstall continued and could report complete with REA still registered. The retain-and-continue exception now applies only to storage REA does not own, classified once by clientSharedRegistrationPaths (shared configuration plus another product's same-scope fallback).
Covers the Greptile P1 on Uninstall.ts. Validation: the new test fails on the pre-fix Uninstall and passes after it; freebuff/zcode/updateMaintenance/bom/dsh suites green; clientLifecycle, clientRegistrationStatus and claudeDesktop keep the same 8 host failures as the pre-merge baseline (symlink EPERM, stale-vs-aligned on Windows).
* fix(update): retain FreeBuff when ZCode reads the shared fallback
Maintenance deduplicated clients on the path each inspection read, so ZCode's fallback view of ~/.agents/mcp.json marked that file "covered" and dropped FreeBuff from the scope. ZCode setup writes ~/.zcode/cli/config.json and deliberately leaves the fallback untouched (docs/client-configuration.md), so the emitted command refreshed only ZCode while the shared row both products load stayed pinned to the old package version. Each setup action rewrites only its own configuration path and no other action rewrites the shared row, so every selected client keeps its own maintenance action; the read-path covered/deduped filter is removed.
Covers the Codex P2 on UpdateMaintenance.ts:104. Validation: the new regression fails on e2fdec384 with {clients:["zcode"]} and passes with {clients:["freebuff","zcode"]}; updateMaintenance 31/31, setupClientConfigCleanupOwnership, cliUpdateAlias, updateOutput, cliStartupImports green; tsc --noEmit, oxlint (0 errors) and oxfmt --check clean; clientRegistrationStatus keeps the same 4 symlink/stale-vs-aligned failures this Windows host has on base e2fdec384 (verified by stashing the change).
🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
* fix(uninstall): preserve completed reads as shared ownership evidence
hasOwnershipEvidence caught every read failure and kept scanning, so a handle-close failure after a successful read — a RegularFileCleanupFailure whose completed outcome still contains the exact shared-file text — was discarded as "no proof". FreeBuff's applicability-unverified storage therefore never got its ownership fact, and once the retained handle blocked later reads, readClientConfiguration downgraded them to retained: inspection passed, removals soft-failed, and runUninstall(true) could purge REA data while the active shared registration remained. The catch now parses the completed outcome for the ownership text while missing or unreadable storage and malformed data keep their proves-nothing semantics, so ownership holds and every later read failure stays blocking.
Covers the Codex P2 on Uninstall.ts:252. Validation: the new regression fails on 4dbfdf1a9 (freebuff reports no blocking failure while the poisoned-scope error lands on claude_code) and passes after; setupClientConfigCleanupOwnership 25/25; the other 21 Uninstall-consumer suites are byte-identical to base 4dbfdf1a9 (61 pre-existing Windows-host failures: symlink EPERM, stale-vs-aligned); tsc --noEmit, oxlint (0 errors, same 6 warnings) and oxfmt --check clean.
🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
* fix(uninstall): keep owned-row proof through removal refusals
hasOwnershipEvidence treated any clientRegistrationRemoval refusal as no ownership: an effective REA-owned rea member shadowed by an earlier foreign duplicate (shadowed_unowned_member), or an owned native row beside a foreign V1 fallback (foreign_legacy_fallback), discarded the already-parsed positive row fact. With FreeBuff unverified and close failures recurring on the preflight/removal rereads, those failures were then downgraded to retained, so runUninstall(true) removed the skill and purged REA data while the active owned row remained. The refusal now carries the owned rows that motivated it — the refusal only refuses the edit — and hasOwnershipEvidence evaluates that fact on both paths; preflight and removal keep refusing the unsafe edit exactly as before.
Covers the Codex P2 on Uninstall.ts:269. Validation: the new regression fails on 225cb214a (freebuff reports no blocking failure while the poisoned-scope error lands on claude_code) and passes after; setupClientConfigCleanupOwnership 26/26; the 21 Uninstall-consumer suites are byte-identical to the recorded 225cb214a baseline (61 pre-existing Windows-host failures: symlink EPERM, stale-vs-aligned); tsc --noEmit, oxlint (0 errors, same 6 warnings) and oxfmt --check clean.
🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
* fix(uninstall): keep unresolved discovery probes from softening shared failures
The discovery probe swallowed every stat failure, so a FreeBuff profile denied with EACCES/EIO was indistinguishable from an absent profile: shared storage was marked applicability-unverified and, when ~/.agents/mcp.json also remained unreadable, the ownership and removal reads softened to retained. runUninstall(true) then removed the skill and purged REA data while an installed FreeBuff could still hold an active registration — the regression even reports complete. Absence (ENOENT) is now the only state that leaves shared paths unverified; any other probe failure leaves applicability unresolved, and shared-config failures stay blocking.
Covers the Codex P2 on Uninstall.ts:318. Validation: the new regression fails on 7572a7082 with status complete and purged data and passes after; freebuffClientConfiguration 13/13; the Uninstall-consumer batches are byte-identical to the recorded baselines (61 pre-existing Windows-host failures: symlink EPERM, stale-vs-aligned); tsc --noEmit, oxlint (0 errors, same warning rules) and oxfmt --check clean.
🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
* fix(uninstall): retain completed configuration reads as removal evidence
A close failure after a completed read discarded the text that read
produced, so an owned registration row the text proves softened to
retained. The completed text now drives removal, keeping shared-config
failures blocking when ownership evidence exists.
Co-Authored-By: Anthropic <noreply@anthropic.com>
* fix(uninstall): stop removal when a cleanup retry leaves a handle unreleased
A completed reread whose handle the cleanup retry cannot release now fails
blockingly instead of serving the text: with no owned row, the soft path
let removeAll purge managed data before close() reported the incomplete
cleanup. The completed text is served only after the retry releases the
handle; an unreleased handle is REA's own failure, never applicability
uncertainty, so it bypasses the retained softening.
Co-Authored-By: Anthropic <noreply@anthropic.com>
* fix(uninstall): keep the ZCode fallback when its profile probe is unresolved
A denied or failed profile probe leaves ZCode's applicability unresolved,
never absent, but the fallback branch dropped the shared row on
applicability alone: with the native config and FreeBuff's profile both
absent, an installed ZCode kept its uninspected registration while
uninstall purged skill and data. Unresolved probes now keep the fallback
in scope with the same fail-closed treatment as other unresolved shared
paths; definitive absence still keeps ZCode out of another product's
primary storage.
Co-Authored-By: Anthropic <noreply@anthropic.com>
* fix(setup): preserve a shared row's enable:false during FreeBuff repair
The shared agents file doubles as ZCode's fallback, so its `enable: false`
binds every reader: doctor now reports the disabled row as stale for
FreeBuff, but following that remediation rebuilt the generic JSON entry
without the flag and silently re-enabled a registration ZCode disabled.
Repairs of shared storage now carry the activation fact forward, while a
client's own native storage keeps its established single-reader repair
contract; the denied-profile uninstall regression also pins the
fallback's new fail-closed inspection beside the denied read.
Co-Authored-By: Anthropic <noreply@anthropic.com>
* test(uninstall): retarget the completed-reread regression to a released retry
The concurrent-writer regression held the reread handle open forever,
which the fail-closed path now serves: it duplicated the unreleased-handle
regression while documenting the opposite contract. The fixture's close
spy gains a failure limit so the cleanup retry releases the handle, and
the test again proves the completed text drives removal of the
concurrently added owned row, with purge verified against the cache
layout purge actually removes.
Co-Authored-By: Anthropic <noreply@anthropic.com>
* fix(doctor): report a preserved shared disable as a manual conflict
Preserving a shared row's enable:false left doctor recommending a setup
run that registrationCurrent correctly treats as already current, so the
scoped doctor returned needs_human forever. The shared-enable
interpretation now has one owner, sharedEnableFlagBinds, used by both
the activation rule and the remediation: a row disabled through the
shared flag reports the manual conflict and states that rea setup will
not change it, instead of an endlessly ineffective repair.
Co-Authored-By: Anthropic <noreply@anthropic.com>
* fix(doctor): keep a native ZCode disable on the setup repair path
The manual-conflict remediation matched a native zcode row through
sharedEnableFlagBinds, but setup drops enable:false on single-reader
storage, so a native disabled registration was told setup would not
change it. The remediation now reads setup's own writer rule,
exported as setupPreservesDisabledActivation: shared storage and the
zcode fallback view keep the manual-conflict message, while a native
row keeps the working rea setup instruction. Regressions pin both
remediations.
🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
* revert: drop the unrelated managed-CIL describe split
The split does not belong to FreeBuff client support and only widens this
PR's diff.
🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
* fix(uninstall): remove the shared agent registration once for both clients
ZCode's ~/.agents/mcp.json fallback and FreeBuff's own configuration queued the
same physical file twice, so ZCode removed the row and FreeBuff then reported
it absent, losing the outcome for the client that owns the file.
Resolve every configuration path to one owner -- the client whose
configuration it is -- and project that single inspection and removal to each
client that loads it. A fallback reader still contributes its discovery probe,
so an installed or unresolved reader keeps the shared file fail-closed.
Discovery also accepted a dangling ~/.config/freebuff-desktop link, which setup
and doctor reject because they follow the target, so an unrelated malformed
shared file blocked every other removal. Resolve the link before treating the
profile as installation evidence, keeping lstat for removal safety.
Also date the FreeBuff release wording against the published 6.7.0 package.
* fix(uninstall): keep a row per client when two clients own one file
OPENCODE_CONFIG can point OpenCode at FreeBuff's configuration file, where
OpenCode writes `mcp.rea` and FreeBuff writes `mcpServers.rea`. Grouping by
filename gave that path a single owner, so uninstall removed only OpenCode's
row, reported that removal for FreeBuff, and could purge managed data with
FreeBuff's registration still active.
Share the decision only through a fallback view of another client's row: every
client that declares the path as its own configuration keeps its own
inspection and removal, and the fallback reader reports each of them.
* fix(clients): preserve discovery and fallback registration identity
Follow product discovery link targets without losing configuration entry
presence or explicitly selected inspections. Associate fallback readers with
the format owner's registration table and project outcomes by owner and path,
so OpenCode removal cannot be reported as a ZCode fallback removal.
All three new boundary regressions fail against the previous source and pass
with the fixes. Validation: check:fast; 217 focused tests passed; the two
changed test files were rerun with 50 passing tests. No installed FreeBuff
client was launched.
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Codebuff <noreply@codebuff.com>
* fix(browser): report an out-of-range script_index as a selection fault
A valid export_web_scripts manifest without the selected script_index was reported as an invalid_format input issue, as if the manifest file were malformed. It is now an out_of_range issue on script_index with the valid bounds, or invalid_value when the manifest has no scripts.
* test(browser): keep the out-of-range index case in its own block
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Passive page capture retained console entries and websocket frames with no
aggregate ceiling. Every sibling retention surface already enforces
WEB_RUNTIME_LIMITS.retainedEventBytes (8 MiB): the observation session's
pre-arm buffer and document-request identities, the browser runtime sources,
requests, and metadata, the V8 inspector's retained metadata, and this same
capture's script metadata, which has its own count-and-bytes budget. A page
that logs a few hundred megabytes of console text, or sends text frames whose
JSON shapes carry hundreds of thousands of keys, grew the capture without
limit until the process ran out of memory; the MCP response budget runs after
the handler, so the growth happened before any typed refusal could exist.
Admit each console entry and websocket frame against the same shared budget
before retaining it. A refused entry is counted through the completeness
tracker's dropped-events vocabulary, which already carries console_events and
websocket_frames kinds and was so far used only for late scripts, so the
inspection reports a truncated inventory with the dropped count instead of
silently overflowing. Console entry bytes measure the retained representation:
the entry's type, url, argument-type strings, and the text capture's retained
bytes, which the handler already computed per entry. A websocket frame retains
its shape summary rather than the payload, and the shape grows with the
payload's own keys, so the frame is charged the serialized size of the shape
it retains. beginAuthorizedFrame resets the counter together with the rest of
the per-frame state.
Out of scope: response metadata and agent hints, whose rows are fixed-shape
facts already bounded per response, and any new setting. The shared 8 MiB
constant is reused as-is.
Validation: RED then green in src/browser/CdpCaptureEvents.eventBudget.test.ts,
where the console case previously retained every oversized entry and the
websocket case retained the over-budget frame; 26 tests across the neighboring
console, websocket, coverage, and completeness suites pass. On this Windows
host the wider browser project has 26 pre-existing failures in three
Electron/WebBundle worker-spawn files, reproduced identically with this change
stashed (26 failed, 644 passed on the pristine base; 26 failed, 647 passed
with it on the same file set). tsc --noEmit over the changed set exits 0,
oxlint reports 0 errors, oxfmt --check is clean, and verify:module-boundaries
passes. No real browser session was driven on this host; the browser CI lane
owns that coverage.
Each retained entry now carries a conservative fixed overhead, and each captured console text value a wrapper cost, so a stream of minimal entries cannot admit millions of objects into the budget when shapes are disabled; a mixed-kind test locks that console entries and websocket frames spend one shared allowance.
Return typed fetch outcomes with completed map coverage when acquisition is
cancelled or a response resource fails. Compose those facts with the captured
inspection as partial bundle Evidence, preserving exclusions, exact abort causes,
and cleanup priority without claiming unperformed script analysis.
Keep complete response-body SHA-256 and byte counts when map interpretation
fails, while preserving compact failure rows and capacity for later valid maps.
Reuse the read owner digest for successful text artifacts and preflight their
JSON representation before parsing. Historical failed rows remain importable.
Exercise controlled HTTP/stream boundaries, CLI SIGINT, and SDK cancellation
with later MCP Evidence retrieval. Share the public cancellation fixture and
use one Result API across consumers. Real Chromium and Windows CLI cancellation
were not verified.
Hash the observed database version with object-store definitions at the CDP boundary. Missing, invalid, or imprecise versions preserve store evidence with incomplete coverage and an explanation. Mark the revised projection so historical store-only hashes cannot create false changes or prove unchanged versions. Preserve comparable metadata and independent storage changes.
Remove the redundant storage comparability wrapper and use each fingerprint's parsed scope instead of interpreting its identity string.
Validation: producer-to-comparison regressions and a compiled CLI to stdio MCP comparison passed, including version-only changes, invalid versions, legacy projections, and complete database additions/removals. Static and docs checks passed. CDP responses were simulated; no real browser runtime was exercised.
* feat(browser): record when scenario requests were observed and how long they took
Request events carry observed_at, the wall-clock time the provider observed
the request start, and finished or failed events carry the browser's
elapsed_ms when it reports one. Both are optional, so earlier captures still
parse, and scenario comparisons leave them out as a committed built-in rule.
The real-Chrome network lane checks every request and finished response is
timed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(environment): time browser scenario records so their state learns a write clock
Scenario records take startedDateTime from when each request was observed and
time from the browser's duration, as HAR entries are timed. A timed scenario
of the book session learns the same clock and serves the same writes as its
HAR; an untimed one keeps record order and learns no clock, as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): document scenario request times and timed scenario records
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): start scenario records at the browser's own request start
The provider observes a request up to tens of milliseconds after the browser
starts it while it reads other responses, so a record starting then and
ending after the browser's duration mixed two clocks. Finished and failed
events now carry the browser's started_at, records prefer it and fall back to
observed_at, and comparisons drop it too. Tests cover the fallback for a
request that never ended or ended twice, a partly untimed capture keeping
record order, comparing an untimed capture with a timed one, and the
real-Chrome lane checks every finished request carries the browser's start
and duration.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): document the browser start time, its fallback lag and older readers
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): record the scenario timing verification
Gates, the real-Chrome lane, prompt-recorded and re-recorded published
scenario captures built with and without their times, the stale published
environments left unrepublished, and the review.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(browser): compare timed completion events in the request-time comparison test
The comparison test's captures had no finished or failed events, so dropping
started_at and elapsed_ms from them went untested. The captures now end one
request and fail the other, the test asserts both carry times, compares them
shifted by seven seconds and against an untimed capture, and a control shows
a changed failure on a timed completion is still reported.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* feat(environment): report what an environment serves with an environment-coverage view
A new inspect_analysis_view kind over capture Evidence. It projects the
state an environment derives with the given template rules and reports
each tool as answered from state, replayed with its view reason, a served
write, or a withheld write with its reason, the records behind it and a
hint about what to record next. The hints cover every write reason, so a
new reason without one fails to compile.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(environment): build an environment bundle in one command
rea environment-build <capture> [directory] inspects a HAR (or reads
capture Evidence), projects the environment manifest from the in-memory
Evidence, authenticates the pair, prepares the state and writes the bundle
environment-export writes. Every bundle file is byte-identical to the
three-step path for the same capture and options.
The report adds the capture and manifest identities, the coverage summary,
warnings (a loopback capture named after its file, credential-named body
members by pointer) and the next command. --dry-run writes nothing. The
default template rules are the four the published environments use.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(environment): cover environment-build and the coverage view
Domain tests check the hint table against the state view's reasons, the
summary, and the view against the environment-state view it summarizes.
Service tests build the fixture capture and compare every file with the
three-step path, a dry run, loopback naming and the refusals. A CLI test
compares a HAR built in one step with the three CLI steps byte for byte,
and an MCP test projects the view from every source kind.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(environment): build the real-agentenv lane's bundle with environment-build
The lane now runs the command users run: it builds the Grafana-shaped HAR
with environment-build and the packed checkout, checks the report, then
runs the bundle with agent-env as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): document environment-build and the coverage view
The agent environments guide opens with the one-step build: the report,
the hint for each withheld write reason, and what a bundle ships. The CLI
guide, the capture guide and the skill reference name the command and the
environment-coverage view.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): harden environment-build after review
- The credential scan walks with its own stack, so a deeply nested body
cannot crash the build, and it also reports credential-named header
names, query names, form body members and string arrays.
- The build checks SIGINT between stages: cancelled before the bundle
stage it writes nothing, during the write it names the written bundle.
- A dry run opens the package tarball, so it refuses what a build refuses.
- Repeated template rules are refused, a HAR saved as .json gets a hint,
file-derived names follow the bundle name rules, `localhost.` counts as
a loopback origin, and the next command is shell-quoted.
- --capture-timeout-ms passes through to the HAR inspection, and a
missing directory is refused before inspection.
- The no-observation hint covers updates and deletes of entities the state
never held; the coverage view states that it does not apply the
serve-time admission checks; the inspect_analysis_view description
names the view.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): state the build's conditions and record its verification
The guides now state when the one-step bundle equals the three-step one,
which headers inspection excludes, what the credential warning scans, how
cancellation behaves, and that the coverage view does not apply the
serve-time admission checks. The spec folder records the census, the end
to end runs, the gates and the review.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): dispatch a pending interrupt before each build stage check
The manifest, state and coverage stages run synchronously, so a SIGINT
that arrived during them was only dispatched once the bundle write had
begun. Each check now yields first, so an interrupt before the bundle
stage cancels the build with nothing written.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(browser): send declared requests from a scenario page with a request step
A capture_browser_scenario step `request` sends one declared HTTP request
through the page's own fetch, with its cookies and session, the way the
application's frontend would, and reads the whole response so the capture
records the transaction like any other. The path must resolve to the start
URL's origin, and the page must be on that origin when the step runs.
Query values and header values may be literals or declared secrets, and a
header may read its value from a cookie or a local storage entry, which is
then redacted from later records like a declared secret. `expected_status`
fails the step on any other status. Scenarios without a request step are
unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(browser): verify request steps in a real browser through to a built environment
verify:browser:request records a write session against a fixture item API
that refuses writes without its CSRF cookie header and stored bearer
token. It checks that every request lands in its own step, that the page
values authenticated each write and appear in no record, header or body,
that an unexpected status fails the step and cancels the rest, and that
environment-build serves the recorded create, update and delete.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(environment): export environment bundles over MCP
export_environment_bundle writes the bundle environment-build writes, from
capture Evidence retained in the session or one capture Evidence file,
into an absolute output directory that must not exist and whose parent
must exist, as extract_firmware and export_web_scripts require; it never
creates the parent or reuses a directory. It returns the same report,
whose schema now lives in the domain so the CLI and the tool share it, and
dry_run reports without writing. A loopback capture without a file needs a
name. The build's refusal messages no longer name CLI flags.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): scan scenario credentials, own the dry-run tarball and serialize Evidence once
A browser scenario's retained header values and the query parameters of its recorded URLs are scanned for credential
names as a HAR's are, through a review projection of its records. A dry run admits and releases the package tarball
through the export's regular-file scope, so a close it cannot confirm keeps its owner, and checks for cancellation
after that last await. Evidence is written into one exactly sized buffer instead of chunks joined by a copy.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): record the review round's checks
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(skill): build agent environments from a prompt with request steps and the bundle tool
Add the environments reference to the skill and install it with setup: record
against a disposable instance, sign in through the UI, write with request
steps and read each write back, check coverage, re-record by withheld reason
and export the bundle. Route it from SKILL.md, name agent environments in the
skill description, and document the request step, the export tool and
recording writes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(browser): split the request lane and header resolution into smaller functions
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* style(environment): format the build spec
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): say when an untimed capture learns no write clock
Coverage of a capture whose records are not all timed, which includes every
browser scenario capture, now states that the state learns no clock, so an
update stamped with a server-set modification time stays unexplained-change
however often it is recorded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(skill): give agents a whole scenario shape with step timeouts
Agents guessed the scenario schema field by field and two hung on steps
without timeout_ms. The reference now starts from a complete scenario with a
timeout on every step, and says a scenario capture cannot serve updates the
application stamps with a modification time.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(browser): keep request steps on their origin and their page values redacted
- Follow no redirects: a redirect fails the step and nothing reaches the
location it names (checked in real Chrome), and check the page's origin
again in the page as the request is sent.
- Send nothing once the step's time has run out or it was cancelled while
header values were read; the in-page deadline is what remains of the step's.
- Keep every page value redacted under a repeated or colliding label, keep a
value's first marker when it is read again, and leave values shorter than
8 characters unredacted rather than rewriting ids, ports and hosts.
- Refuse a header name given twice, User-Agent (Chrome drops it) and line
breaks in literal values and prefixes; fail before sending when a secret or
page value holds them.
- Append query entries after the path's literal query without re-encoding it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): document request step limits and word bundle refusals for both surfaces
Document redirects, timeouts, page value redaction, HttpOnly cookie sources
and header refusals for request steps. Refusals shared by environment-build
and export_environment_bundle no longer name CLI flags, and the retained
loopback refusal no longer prints an empty file name. The skill reference
asks for read-backs after each write, shows the coverage view's source, names
package_tarball for unreleased builds and the browser path a scenario ships.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): scan every form occurrence and MIME-less scenario bodies
The build's credential scan collapsed form bodies with Object.fromEntries, so a
repeated field kept only its last occurrence: `api_key=secret&api_key=` warned
about nothing although the Evidence ships the value, and repeated values were
undercounted. Form fields are now scanned as ordered name/value pairs, as
headers and query strings already are.
The review projection of a capture_browser_scenario dropped a request body
recorded without a media type, so the scan never saw it. Review keeps such a
body's text with an empty media type; the manifest projection still reads only
bodies it can classify.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(browser): close request step edge cases from the second review
- Append query entries correctly after a path ending in "?", "&" or "#".
- Refuse 3xx expected statuses, which an unfollowed redirect can never
report, and say a redirected write may already have been applied.
- Redact a page value also as fetch sends it, without surrounding spaces
and tabs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): document page value body costs, redirects and MCP dry runs
Document that reading a page value shrinks the bodies a capture keeps under
the default allocation limit, that redirected writes may have landed, that
the MCP dry_run still checks its output directory and which refusals name
the tool, and keep the .har hint in the HAR-as-JSON refusal for both
surfaces.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): record the prompt-to-environment verification
Gates, the real-browser lanes, five environments built from prompts by an
agent on disposable VMs and evaluated with two models, and the two reviews.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(skill): count rollback failures from the managed skill files
The environments reference made the skill eight files, so the recovery tests now derive the counts they expect from
MANAGED_SKILL_FILES instead of fixing them at six and seven.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(adb): restore the portable Java float pattern a merge rewrote
A lint autofix during a merge unescaped the hyphens in javaFloatValuePattern, which is invalid in a v-mode character
class, so the advertised start_adb_activity schema failed to compile in every JS regex mode. The file now matches main.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(browser): show the request step inside a whole scenario
Every JSON example in the browser scenario guide is a capture_browser_scenario input the catalog test parses, so the
request step example now sits in a launch scenario instead of standing alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): scan structured HAR form params and the request side of unattributed scenario transactions
A HAR request recorded with postData.params and no text is scanned by those parameters, each occurrence counting, and
the review projection keeps the request side of a scenario transaction whose response cannot be attributed, so the
credential warning covers what evidence.json ships in both cases.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): validate build paths and read dry-run tarballs
* fix(environment): report bodies too large for the credential scan
Text bodies over the scan's 4 MiB limit, including a response decoded only as a prefix, are reported as a
credential-scan-incomplete warning by record and side, never by content, so an absent credential-named-members
warning no longer reads as a complete check. The spec and agent-environments guide also state the path validation and
dry-run tarball read that e52ae1cd added.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(verification): release cold source transform services
* fix(javascript): retain capture context along covered mutation paths
* fix(environment): guard request dispatch and report export limits
* fix(environment): admit serialized exports and scan HAR form fallbacks
* fix(environment): admit in-memory build inputs against heap headroom and scan SaveHar null-text forms
The inspected HAR's Evidence and the projected manifest are checked against live heap headroom before they are
serialized, with the same check the file route uses, so a capture too large to parse and seal is refused as
resource_constraint before anything is written. A request whose postData text is SaveHar's null is scanned by its
structured params.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): measure the credential scan limit in UTF-8 bytes
A body counts against the 4 MiB limit by its UTF-8 length, not its UTF-16 code units, so multibyte text over the
limit is reported as unscanned instead of parsed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(browser): send request steps only before their deadline and keep page reads on the start origin
The page receives the step's absolute deadline and a one-second start window, so a busy page that runs the request
late, or after the step is cancelled, sends nothing; local storage is read only on the start origin; the response is
drained without keeping its body; an unparseable path is a field issue; retained Evidence is refused for heap headroom
before it is serialized; and the skill says how to reach server-assigned ids.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* style: format integrated export admission
* fix(javascript): erase inline type imports before binding values
* fix(browser): run the request lane under the shared verifier command owner
The browser request lane runs each CLI call through createPackageVerifierCommands on a private runtime root, so a
timeout, output overflow or failure stops the CLI and the browser it launched before the root is removed, and a failed
stop is reported with the root retained. The verification record states the lane's current counts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): scan scenario content by event and read bodies by their media type
A browser scenario's credential scan reads every retained network-content event, including a response attached
mid-flight and a transaction no record represents, with event pointers, and each transaction URL's query once; the
review record projection it replaces is removed. A body is parsed as a form or as JSON by its recorded media type,
falling back to its shape, so valid forms such as `password=x&` are scanned and sentences are not read as forms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Remove duplicate MCP defaults, the fixed CDP deadline message, and the canned Windows authority reason after verifying their only consumers were tests.
Pin independent expected budget facts, exercise active default and selected CDP deadlines, and verify exact authored native diagnostic propagation. Preserve malformed-input, cancellation, omission, and authority assertions. Verified 222 focused cases, 9 README cases, build, static checks, docs, discovery, and duplication at 3.58%.
CDP joins repeated response header field lines with LF. Only the Link projection split them, so a repeated Content-Security-Policy lost the first directive of each later policy, and a repeated Referrer-Policy reported no policy. Header normalization now keeps each field line, CSP parses every line and comma-separated policy, single-valued fields combine lines with a comma, and Link uses the shared lines.
Resolve the filesystem-selected manifest parent once before deriving script paths and reuse the shared containment rule. Preserve selected manifest spelling and capture metadata while retaining source integrity, symlink refusal and cleanup. Cover differing and identical lexical siblings through both artifact consumers and public CLI/MCP parity.
Read screenshot identity from the selected frame-tree root instead of a
filtered inventory that could substitute a child. Reuse the frame record
parser for initial and event identities, rejecting malformed roots before
capture while preserving known origin refusals.
Share explicit origin-scope diagnostics across CDP workflows without
changing policy or conflating startup uncertainty, document replacement,
worker protection, and Inspector restrictions. Document the recovery and
make screenshot navigation fixture ordering explicit.
Validation: 577 focused checks passed across 13 files; one platform-specific
invalid-byte-filename case skipped. Build, typecheck, lint, formatting,
generated documentation checks, README conformance, and discovery passed.
No full suite or live browser/provider execution.
Decode base64 JSON body bytes at the shape owner without replacement characters. Keep incomplete request correspondence unknown and use explicitly complete authored inventories for positive comparison controls. Repair semantic truncation and storage fixtures to satisfy their owning contracts without weakening assertions.
morluto