Train 1109: #534 C# bind account, #562 index_state contract, #546 Vue SFC, #531, #557, schema 93 cost #564

Merged
buildagent merged 63 commits from integration-1109 into master 2026-11-10 10:34:56 +01:00
Member

Integration train merging the codex-f1…f5 fix branches, plus the audited #534 re-bless and the record re-records.

  • #534
    • C# member and namespace proofs: namespace_type provenance, a refused search on failed linearization, the COALESCE bypass removed, and a shared alias veto. Readiness is 28/28 and the bind diff is zero.
    • baseline.json is re-blessed with a bind account (eefa9b6f).
  • Cost: schema 92→93 adds covering/partial indexes. Every pinned repo is below its old band except cs-dapper at +6.6%, which is attributed statement by statement to the #534 proofs.
  • #562: the index_state map lookup is total (fixes a panic). The index_incomplete contract is refined: a symbol miss during resolution_pending is a measured absence.
  • #546: the shipped Vue SFC fix; plugin add previews with the languages the packages claim.
  • #531: py-django tier-3 −42. 56 removed binds, all wrong; 14 added, all correct.
  • #557: payload ratchets; the startup reserve is 627 ≥ 622.
  • Records: re-recorded in 60368c0d with reasons: stage, cost, tier3, host package lifecycle, Ruby package cost, and the agent bench. In the bench, cs-dapper fallback went 25→15, all from who_calls.SetupCommand; recall is unchanged.

Local full gate run on 60368c0d, on an idle machine: all 44 legs green. That covers fmt, clippy, docs, the workspace and daemon legs, precision 7/7 with zero phantoms, the corpus suites, all package legs, tier3, mixed_load, the acceptance suites, plugin_path_cost and agent_bench.

🤖 Generated with Claude Code

Integration train merging the codex-f1…f5 fix branches, plus the audited #534 re-bless and the record re-records. - **#534** - C# member and namespace proofs: namespace_type provenance, a refused search on failed linearization, the COALESCE bypass removed, and a shared alias veto. Readiness is 28/28 and the bind diff is zero. - `baseline.json` is re-blessed with a bind account (eefa9b6f). - **Cost:** schema 92→93 adds covering/partial indexes. Every pinned repo is below its old band except cs-dapper at +6.6%, which is attributed statement by statement to the #534 proofs. - **#562:** the index_state map lookup is total (fixes a panic). The `index_incomplete` contract is refined: a symbol miss during `resolution_pending` is a measured absence. - **#546:** the shipped Vue SFC fix; `plugin add` previews with the languages the packages claim. - **#531:** py-django tier-3 −42. 56 removed binds, all wrong; 14 added, all correct. - **#557:** payload ratchets; the startup reserve is 627 ≥ 622. - **Records:** re-recorded in 60368c0d with reasons: stage, cost, tier3, host package lifecycle, Ruby package cost, and the agent bench. In the bench, cs-dapper fallback went 25→15, all from `who_calls.SetupCommand`; recall is unchanged. Local full gate run on 60368c0d, on an idle machine: all 44 legs green. That covers fmt, clippy, docs, the workspace and daemon legs, precision 7/7 with zero phantoms, the corpus suites, all package legs, tier3, mixed_load, the acceptance suites, plugin_path_cost and agent_bench. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Codex <noreply@openai.com>
Nested worktrees answered from the parent tree when its configuration marker was found before the worktree Git boundary. Walk upward once so the nearest root marker wins.

Co-Authored-By: Codex <noreply@openai.com>
A .code-index.toml above a nested worktree or clone captured it, so
agent worktree sessions answered from the parent tree. One upward walk;
the innermost marker wins.

Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Codex <noreply@openai.com>
Share owned transaction cleanup across generation changes, FTS maintenance,
and the write-access probe so a failed COMMIT cannot strand the connection.
Preserve caller-owned transactions and cancellation-safe rollback.

Add a production COMMIT registry and a real SQLITE_BUSY commit regression
with durable connection reuse; verify both tests reject their mutations.

Co-Authored-By: Codex <noreply@openai.com>
Highest version per package id; equal highest versions under several
digests defer instead of picking one; no older row is enabled when the
newest is not licensed.

Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
One owned-transaction helper (busy::with_owned_transaction) returns the
connection to autocommit after a BEGIN, body or COMMIT failure; every
hand-written production transaction goes through it, and a registry
test rejects a bare COMMIT outside it.

Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Point every agent at the zero-fallback epic from the files they read at
startup (CLAUDE.md for Claude, AGENTS.md for Codex and others).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Codex <noreply@openai.com>
A shared daemon no longer dies when the client that started it is
signalled; leashed test runs keep the inherited group via
COSI_TEST_LEASH_GROUP so their cleanup still reaps.

Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Codex <noreply@openai.com>
An ignored suite measures how often agents fall back to the shell on
indexed code, per Claude/Codex transcript; indexed paths are decided by
the production walk policy (walker::Coverage).

Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Share enclosing-namespace precedence and import ambiguity selection across
simple type references, inferred receiver origins, and inherited lookup.
Gate the new selection through the C# partial-type profile and retain
existing decisions for other profiles and qualified spellings.

Check type arity before namespace shadowing and use stable source order
for new partial-type targets. Add namespace collision and ambiguity
fixtures plus precision decoys.

Co-Authored-By: Codex <noreply@openai.com>
- corpus gates and precision_gate install the checkout packages the
  .digest records name (the bytes that ship), not the first frozen pin;
- corpus_bind_diff: reusable bind-level diff joined on
  (path, line, col, kind, occurrence), split by resolved_by;
- C# simple type names: enclosing namespace chain before using-imports,
  ambiguous imports bind nothing (26 wrong shipped binds fixed).

corpus_ratchet is expected red until the re-bless, which waits for the
Rust index-receiver regression (#540); see the bind account on #534.

Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Capture schema at pool opening and serve health outside query workers.
Leave optional activation identity unreported on liveness rather than
cache a generation that another process can promote; query endpoints
retain the current database identity.

Bound read checkout with typed BUSY, keep request refusals distinct from
connection-cap refusals, and cancel queued dispatch when its peer closes.
Preserve bounded pipelined input while watching for EOF.

Exercise twenty real-daemon handshakes with every read connection held,
verify permit/slot recovery, and run the timeout, cancellation and health
mutations against their intended tests.

Co-Authored-By: Codex <noreply@openai.com>
Health needs no pool connection or SQL and takes the socket-runtime fast
path; ReadPool::get waits at most 1 s and refuses with typed BUSY;
queued requests whose peer has gone never take a connection.

Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Reuse the bounded git blob reader for historical path/range reads.
Attach clipped, budgeted hunks to changed symbol rows, including deleted
base-side bodies, and carry source end lines through the outline RPC.
Exercise both transports, wire skew, default reply fields, and mutations.

Co-Authored-By: Codex <codex@openai.com>
Worktrees created from the same branch share HEAD, so head,
uncommitted_changes and sibling_worktrees alone cannot tell an agent
whether a reply came from its worktree or from the main checkout (#553).

When sibling_count > 0, emit root on each RootProvenance entry via
display_root: the repository's common checkout (first entry of git
worktree list --porcelain, or any external worktree) carries its
absolute path once, while a worktree nested under the common checkout
is shortened relative to it (.claude/worktrees/<name>). Lone checkouts
(sibling_count == 0) and unavailable trees omit the key for the same
per-reply budget reason SiblingWorktrees::Measured(0) is silent.

Co-Authored-By: Gemini <noreply@google.com>
Worktrees share HEAD, so a reply could not say which tree answered.
The root is now disclosed when a sibling worktree exists, relative to
the common checkout for nested worktrees (+0 B on single checkouts).

Co-Authored-By: Gemini <noreply@google.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Measured on search_symbols: envelope=minimal + concise carries the same
fields at about half the bytes; the option was only named in the last
line of the server instructions, which clients truncate.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Reuse the review's structural-container classification to exclude listed
child spans from container hunks and disclose hunk_in_children. Spend the
shared budget on leaves first without changing serialized row order.

Test unique line ownership, containers with no own text, and budget pressure through
both tools. Run full-span and containers-first mutations against the test.

Co-Authored-By: Codex <codex@openai.com>
read_code takes rev=<commit-ish>; changed_symbols/review_diff take
include_hunks, each hunk owned by the innermost changed symbol under the
existing token budget. A review can stay inside code-index.

Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Preserve index, try and parenthesized receiver spelling without taking a call argument or field suffix as its type. Keep ordinary expression capture unchanged, grade the macro facts with an independent grammar oracle, and rebuild the Rust package as 0.1.8.

Co-Authored-By: Codex <noreply@openai.com>
assert_eq!-style macro token trees cleared the bracket receiver of
msgs[0].unwrap_begin(); 30 correct ripgrep binds return, rust-analyzer
is byte-identical.

Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Add SessionEnvelope per-session state on CodeIndexServer so each (json_pointer, prose_text) pair ships in full on its first appearance in an MCP session and is replaced by semantics_ref: "code-index://docs/envelope" on subsequent replies while keeping every field and value.
- Omit answer_provenance when unchanged since the previous reply in the session and not abnormal (daemon_restarted, daemon_build skew, unavailable/unmeasured tree state, index_state reconciling/unmeasured, or measured-absence error).
- Support envelope: "full" per call and CODE_INDEX_ENVELOPE=full env override; move the envelope hint to the first line of build_instructions.
- Expand code-index://docs/envelope to cover every semantics key.
- Replace FINDING 5 default-envelope test with session deduplication and provenance-change tests.

Co-Authored-By: Gemini <noreply@google.com>
Session envelope by default: a (pointer, prose) pair ships once per MCP
session and is then replaced by semantics_ref; answer_provenance ships
only when it changed or is abnormal. envelope="full" per call and
CODE_INDEX_ENVELOPE=full restore full bytes. Measured -46% on
prose-bearing replies.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
agy-557 dogfood D1: the blanket rule made read_code and project_overview
refuse the first call. Also note the session envelope is now the default.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Codex <noreply@openai.com>
Both were integration regressions from earlier merges in this train.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Gemini <noreply@google.com>
# Conflicts:
#	crates/daemon/src/offer.rs
Reject recorded Rust receiver contradictions, undeclared JS/TS runtime
globals, and name-only cross-crate decisions. Preserve shared mod trees
and receiver proofs. Keep evidence-less std-surface refusal disabled.

Drop the impl-self import guard: its audit loses correct factory and
iterator calls. Nine-repository comparison retains only 31 audited wrong
removals, with zero additions or retargets; live cross-test flush binds
are also gone. Fixtures, mutations and precision gate cover the subset.

Co-Authored-By: Codex <noreply@openai.com>
Loss-free subset (operator decision): recorded receiver contradiction,
undeclared JS/TS ambient globals, independent Cargo crate roots. Bind
diff on all 9 repos: 31 removals, all audited wrong; 0 additions. The
evidence-less std-surface refusal stays off until receiver facts exist.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Share one normalized census between activation offers, extension coverage,
and startup auto-enable. Invalidate it on reconciliation, promotion, and
watcher path-set events while retaining it across content edits.

Prove reply reuse and watcher invalidation with a walk counter, and verify
walk-only extension counts agree in the same payload. Both regression
mutations were run and rejected by their intended tests.

Co-Authored-By: Codex <noreply@openai.com>
agy-546 (4a971b16) plus codex-546r (072e4125): the census is computed once
per (generation, file-set epoch) and feeds offers, coverage and startup
activation alike; no walk per request.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Keep recorded type arity in base-only proofs and resolve nested base names
through the proven bases of their enclosing declarations. Preserve partial
type identity and bounded, fail-closed traversal instead of trusting a
provisional name-tier target.

Cover base.M() beside a generic namesake and an inherited unqualified call
through a nested generic base, with precision positives and decoys. The
nested-lookup mutation fails both the fixture and the C# precision probe.

Shipped/shipped corpus binary diff against the integration snapshot adds
three correct Dapper binds and changes no binds in the other eight repos.
Fmt, workspace all-target clippy, member and namespace tests, resolver unit
tests, and all seven precision gates pass with zero phantoms.

Co-Authored-By: Codex <noreply@openai.com>
Restores the two correct inherited-member binds the shipped C# package
lost (Dapper SqlMapper.Identity.cs:51, SqlCompactDatabase.cs:38) plus one
more (:44). Bind diff on all 9 repos: +3 in cs-dapper, all correct; no
other movement.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Route path reads through project-root validation and the bounded live reader without requiring an index handle or file row. Disclose missing or stale rows and retain proven coverage refusals. Qualify incomplete-index lookup misses in the shared reply layer, with lifecycle state and available progress.

Register the lookup refusal vocabulary and document its contract. Cover startup reads, incomplete lookup misses, and ready-index misses on both e2e legs, with each required mutation observed red.

Co-Authored-By: Codex <noreply@openai.com>
read_code path/range targets are served live in every index state with an
index_pending disclosure; walker, eligibility, content and size checks
still refuse. Lookup misses on a non-ready index become a retryable
index_incomplete at the shared reply layer; ready-index misses keep their
verdict.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Replace C# namespace global maxima with per-file and per-type histograms, scope global imports to their projects, and price tier-3 reachability by directory and import edges. Keep budgets unchanged and retain one degradation verdict and warning across repeated substages and passes.

Co-Authored-By: Codex <noreply@openai.com>
Keep the existing target of a base reference unless positive same-scope proof replaces it. Carry proof completeness separately from a completed external-type lookup, so missing import facts cannot become negative inheritance evidence. Preserve unknown member decisions and incomplete placement checks without weakening the proven-external refusal.

Co-Authored-By: Codex <noreply@openai.com>
csharp_partial no longer degrades on cs-dapper (estimate 9.49e9 -> 2.10e6,
budget unchanged, +2.18% VM). Running it to completion exposed 163 correct
binds withdrawn when an unrelated file lacked import proof; base targets
now fall back to the existing bind. Bind diff vs 40f9e5c7 on all 9 repos:
+4 audited correct, 0 removed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Movement vs the master baseline: cs-dapper resolved +1685, rust-ripgrep +52,
ts-zod -207, js-express -42; no stage degraded on any pin. Bind-level
account (joined on path,line,col,kind,occurrence, split by resolved_by):
every audited removal is a wrong bind; ~175 audited additions and all 32
dapper retargets are correct; the 2 lost and 163 withdrawn correct dapper
binds are restored. Full account on #534.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Measure whole-file refusal spans against the pending generation's accepted size so source edits cannot bypass the producer-loss gate. Keep delegated component scoping intact.

Measure overview disclosure prose in full, retain each delivered resolve interpretation, and price indexed and walk-only populations separately. Pin the expanded Rust/TypeScript fixtures exactly. Restore the overview reserve through truthful, shorter coverage prose and fund startup growth by sharing the freshness description.

Validated parent/merge boundaries, seven mutation cases, fmt, workspace clippy, both MCP legs, and 336 binary unit tests. The requested MCP --lib command has no library target.

Co-Authored-By: Codex <noreply@openai.com>
Separate simple namespace type provenance from qualified FQN binds, check
private nested owners' generic arity, and refuse inherited lookup when no
hierarchy order exists. Bound bridge invalidation by current and stale mapped
names so shipped C# to XAML bridges do not force unrelated outline edits full.

Register SQL reader stances, document manual bind-diff waivers in both CI
registries, pin the audited fixture ref census, and support comparing separately
built resolver binaries. Addresses train regressions after #531 and #534.

Validation: fmt; workspace/all-target clippy with warnings denied; requested
release tests; 11 metamorphic tests; 20 bridge tests; precision 7/7 with zero
declared-decoy phantoms. Nine corpus repositories have zero added, removed or
retargeted binds; 801 provenance labels change. Stage drift is reported without
re-recording any corpus baseline. Each fix was checked with a failing mutation.

Co-Authored-By: Codex <noreply@openai.com>
Read optional index state without indexing a missing map key. Reproduce
incomplete batch entries with no outer state through the annotation path.
Use original error subjects to name batch project state while preserving
caller query labels.

Assert retryable index_incomplete on lookup waits while retaining daemon,
convergence, lifetime and timing guards. Isolate CLI fixture package stores
and request the full envelope when grading progress interpretation text.

Refs #562 and #554.

Co-Authored-By: Codex <noreply@openai.com>
Index reference populations and local declaration probes, seek nearest type
positions without per-receiver sorts, and replace tier-3 autoindexed CTEs with
a keyed directory histogram and existing edge/bucket lookups. Preserve the
estimate's multiplicity and keep candidate-name probes linear on hub files.

Schema 93 carries the indexes. No corpus or cost baseline is changed.
Corpus ratchet and precision pass; cost lower floors, residual C# VM work,
schema-conditioned records and additional train test failures remain red.

Co-Authored-By: Codex <noreply@openai.com>
Include newly added package languages in the activation preview and recorded language decision when automatic maintenance permits it. Preserve standard-language decisions, automatic state files, and maintenance opt-outs. Add a packages-only Vue script regression test.

Keep symbol-name absence authoritative after parsing while reference resolution is pending. Preserve incomplete-index refusals during parsing and for reference/graph queries, with a state/tool matrix regression test. Isolate acceptance fixtures from prior approvals and grade region-only SFC refusal.

Validation: both release-shape acceptance scripts pass; fmt and workspace all-target clippy pass; requested e2e tests pass under ci and ci-timing on both legs. Both new regression tests caught their executed mutations. Broader crate suites retain additional train reds, recorded in the worker report without modifying corpus records.

Co-Authored-By: Codex <noreply@openai.com>
Keep namespace import requirements from replacing inheritance proof; reject alias-vetoed namespace types and retry bridge imports by imported member. Audit package parity differences and verify nested-static targets independently.

Co-Authored-By: Codex <noreply@openai.com>
Avoid project-wide invalidation scans before bound targets or resolver health
exist, while retaining warm all-unresolved retries. Drive base and namespace
call readers by reference keys instead of broad admission/NULL-target plans.
Move the written-type SQL registry stance to its actual file scope.

C# bind exports are byte-identical across 23,287 refs and 5,842 resolved binds.
Cold attribution falls from 30,871,006 to 29,049,192 VM steps. The cost gate
still rejects the remaining C# increase and earlier down-moves; no baseline
is blessed or edited.

Co-Authored-By: Codex <noreply@openai.com>
corpus: re-record stage, cost, tier-3, package and bench records after the #534 account
Some checks failed
CI / cargo doc (intra-doc links) (pull_request) Successful in 8m38s
Package CI / package-corpus (svelte) (pull_request) Successful in 2s
Package CI / package-corpus (timeline) (pull_request) Successful in 2s
CI / cargo deny (pull_request) Failing after 10m28s
Package CI / package-corpus (vue) (pull_request) Successful in 2s
Package CI / package-corpus (xaml) (pull_request) Successful in 2s
CI / cargo clippy (pull_request) Successful in 13m5s
Package CI / package (rust) (pull_request) Successful in 12m32s
Package CI / package (php) (pull_request) Successful in 14m1s
CI / cargo test (abi, 32-bit + wasm32) (pull_request) Successful in 17m27s
CI / cargo check (MSRV 1.99.1) (pull_request) Successful in 18m39s
CI / cargo test (pull_request) Has been skipped
CI / cargo test (daemon transport) (pull_request) Has been skipped
Package CI / package (ruby) (pull_request) Successful in 16m34s
CI / cargo check (windows-gnu) (pull_request) Successful in 20m8s
Package CI / package (python) (pull_request) Successful in 17m18s
Package CI / package (javascript) (pull_request) Successful in 22m33s
Package CI / package (typescript) (pull_request) Successful in 22m54s
Package CI / package (csharp) (pull_request) Successful in 23m19s
Package CI / package-corpus (php) (pull_request) Successful in 28m49s
Package CI / package-corpus (rust) (pull_request) Successful in 30m6s
CI (Windows) / fmt + clippy + build + test (windows) (pull_request) Failing after 35m18s
Package CI / package-corpus (python) (pull_request) Successful in 31m59s
Package CI / package-corpus (javascript) (pull_request) Successful in 32m7s
Package CI / package-corpus (typescript) (pull_request) Successful in 28m6s
Package CI / package-corpus (csharp) (pull_request) Successful in 34m3s
Package CI / package-corpus (ruby) (pull_request) Successful in 37m12s
Package CI / package overhead budgets (nightly, one package at a time) (pull_request) Has been skipped
CI / OSS corpus (tier 1) (pull_request) Successful in 59m21s
CI / Plugin path cost + pool throughput (nightly) (pull_request) Has been skipped
60368c0ded
Each record carries its own reason, written by the bless writer:
- stage-baseline: rule redistribution of the audited #534 re-bless (sum(rule.*) = resolved again)
- cost-baseline: schema 93 indexes took every repo below its band; cs-dapper +6.6% vs master
  attributed statement by statement to #534's C# proofs (codex-f4 ledger)
- tier3-baseline: py-django -42, all 56 removals audited wrong, 14 additions correct (#531)
- host-package-lifecycle, ruby_package_cost: re-measured under schema 93
- bench ratchet: cs-dapper correct_via_fallback 25 -> 15 = dapper.who_calls.SetupCommand,
  10 callers now resolved binds (measured on master e3ec6ead with its own binary)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
cargo deny fails the advisory on every rmcp before 3.9.0. The affected
Streamable HTTP client features are not built here, but the check is by
version. The typed skills handlers this release introduces follow in the
next commit.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014XLNPtTbPwnHBYfKMb3f1W
Merge codex-g1: typed rmcp skills handlers and three Windows test fixes
All checks were successful
CI / cargo doc (intra-doc links) (pull_request) Successful in 8m15s
Package CI / package-corpus (svelte) (pull_request) Successful in 4s
Package CI / package-corpus (timeline) (pull_request) Successful in 2s
CI / cargo deny (pull_request) Successful in 9m27s
Package CI / package-corpus (vue) (pull_request) Successful in 5s
Package CI / package-corpus (xaml) (pull_request) Successful in 2s
CI / cargo clippy (pull_request) Successful in 13m34s
Package CI / package (rust) (pull_request) Successful in 12m38s
Package CI / package (php) (pull_request) Successful in 13m45s
CI / cargo test (abi, 32-bit + wasm32) (pull_request) Successful in 17m28s
CI / cargo check (MSRV 1.99.1) (pull_request) Successful in 18m28s
Package CI / package (ruby) (pull_request) Successful in 16m53s
CI / cargo check (windows-gnu) (pull_request) Successful in 20m17s
Package CI / package (python) (pull_request) Successful in 17m21s
Package CI / package (typescript) (pull_request) Successful in 23m6s
Package CI / package (javascript) (pull_request) Successful in 23m14s
Package CI / package (csharp) (pull_request) Successful in 23m42s
Package CI / package-corpus (php) (pull_request) Successful in 29m36s
Package CI / package-corpus (rust) (pull_request) Successful in 26m20s
Package CI / package-corpus (python) (pull_request) Successful in 33m12s
Package CI / package-corpus (javascript) (pull_request) Successful in 33m41s
Package CI / package-corpus (typescript) (pull_request) Successful in 29m22s
Package CI / package-corpus (csharp) (pull_request) Successful in 35m50s
Package CI / package-corpus (ruby) (pull_request) Successful in 35m37s
Package CI / package overhead budgets (nightly, one package at a time) (pull_request) Has been skipped
CI (Windows) / fmt + clippy + build + test (windows) (pull_request) Successful in 46m45s
CI / OSS corpus (tier 1) (pull_request) Successful in 1h4m41s
CI / cargo test (pull_request) Successful in 54m56s
CI / cargo test (daemon transport) (pull_request) Successful in 6m13s
CI / Plugin path cost + pool throughput (nightly) (pull_request) Has been skipped
b644d7d6b7
rmcp 3.9.0 routes skills/list and skills/get to typed handlers, so the
custom-request hook no longer saw them. The wire result is unchanged.
Windows: the provenance root is rendered with forward slashes, the
fallback-rate classifier normalises separators and the home directory,
and the test-seams daemon builds into a private target directory.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014XLNPtTbPwnHBYfKMb3f1W
buildagent referenced this pull request from a commit 2026-11-10 11:41:13 +01:00
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
h-dv/code-index!564
No description provided.