Skip to content

fix(acp): stop asserting harness-mounted MCP servers absent - #14614

Merged
bolichen97 merged 1 commit into
mainfrom
fix/mcp-ref-guard-claude-harness-mounted-servers
Sep 28, 2026
Merged

bolichen97 merged 1 commit into
mainfrom
fix/mcp-ref-guard-claude-harness-mounted-servers

Conversation

@pepmach

@pepmach pepmach commented Sep 28, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: A harness that mounts MCP servers from its own configuration is not told by Crew that those servers are absent: the unresolved-@server warning says only what Crew can see.

With agent.acp_backend = "claude", kiro_crew.acp.mcp_ref_guard logs its warning with the closing sentence "Those tools are absent from the session with nothing else to say so". Claude Code mounts its own user- and project-scope mcpServers and its plugins beside the array Crew sends, and none of those ever appear in Crew's session/new array. So servers that were mounted and answering in the session were listed as absent, on one line with servers that really were missing (the kirocrew-* servers only travel on Crew's array). An agent that read the line told the user those tools were unavailable, and the user had to correct it. codex-acp has the same shape on record: it merges Crew's array on top of what ~/.codex/config.toml declares, and kiro-cli loads the global ~/.kiro/settings/mcp.json into every agent whatever the spec declares. No backend's wire array (or spec) is provably the session's whole MCP surface.

Why it matters

The line mixes a true finding with a false one and words both as certain. Whoever is diagnosing a broken session, a person or an agent, cannot tell which entries to believe, and the definitive wording sends them the wrong way.

Not a goal

  • Computing the exact list of harness-mounted servers by reading ~/.claude.json, a project .mcp.json, ~/.codex/config.toml or a plugin set. That makes a harness's own configuration file an input to Crew's detector, a seam decision the triage excluded on purpose, and it would still miss what plugins mount. The list stays what the wire proves; only the verdict sentence changes.
  • Changing the detector's satisfaction rule (unresolved_server_refs and ACP_BACKENDS_SPEC_SERVERS_OFF_WIRE are untouched): Crew cannot mark those refs satisfied without reading the harness's configuration.
  • Splitting the list into "really absent" and "may be inherited" buckets.
  • The array withholding for concurrent claude sessions (claude backend: concurrent dashboard chats sharing the default workspace get no Crew MCP tools (settings.local.json seed has a single live owner) #13143), which is what exposed this.

What changed (motivation → approach → change)

Piece What it does
src/kiro_crew/acp/mcp_ref_guard.py The warning's leading clause claims only what Crew can assert ("agent-spec tool refs name no MCP server in Crew's projection for this session"), and its closing sentence is the hedged one on every backend: Crew's projection delivers none of the listed servers, and the harness may mount a same-named server from its own configuration, so a listed ref may still be served and the line cannot tell which. Fields and the mirror remedy are unchanged.
src/kiro_crew/cli_doctor.py The static half of the same detector (_doctor_unresolved_mcp_refs, no-mirror branch) said "absent from its sessions with nothing to say so"; it now carries the same hedged sentence as the runtime line, unconditionally. The broker-stub caveat and the registry pointer are unchanged.
test/test_acp_mcp_ref_guard.py, test/test_cli_doctor.py The full-line test pins the new leading clause and the hedge on every known backend and asserts the old "this session receives" clause is gone; the redaction, alphabet-rebuild and call-site tests are unchanged and still pass. A doctor test pins the hedge on the no-mirror branch for a wire-fed and an off-wire backend alike, with neither "absent from its sessions" nor "nothing to say so".
docs/system-specs/modules/agent-host-contract.md The detector paragraph records what the line claims and why the claim is the same everywhere; its premise now matches its own evidence (a harness Crew reaches only through the array it is sent is satisfied, as far as Crew can see, only by that array), and it notes the doctor row carries the same verdict.
src/kiro_crew/agent_sdk/backends.py, src/kiro_crew/agent_sdk/mcp_refs.py, src/kiro_crew/acp/client.py, src/kiro_crew/acp/harness/codex.py, src/kiro_crew/providers/mirrors/claude_code.py, src/kiro_crew/docs/agent-spec-fields.md, docs/system-specs/modules/claude-code-provider.md Every comment and doc sentence that called Crew's array the session's "entire" or "whole MCP surface" now says what is true: the array is the only channel Crew has onto the session, and a wire-fed harness may mount servers from its own configuration beside it (grep -rni "entire MCP surface|whole MCP surface" src docs finds none).

The symptom is a sentence, not a list. "Absent" is the strong claim, and the detector cannot back it anywhere: it reads the wire (or, on kiro-cli and KAS, the spec) alone, by design, and cannot tell a server the harness mounted itself from one nobody mounted. The hedge therefore applies to every backend, not only the claude session the issue reports, because no backend's wire array is provably its whole MCP surface: kiro-cli's own record in this repository (src/kiro_crew/apps/bridges.py, acp/mcp_session_report.py) is that the global ~/.kiro/settings/mcp.json is loaded into every agent with no off switch. The line says what the detector knows and stops there: "Crew's projection delivers none of those servers; the harness may mount a same-named server from its own configuration, so a listed ref may still be served and this line cannot tell which. The harness still works." The fields and the mirror remedy that follows are unchanged, and the leading clause now says the same thing the verdict does: the ref names nothing in Crew's projection, not nothing the session receives. The line does not claim that such a copy escapes the spec's tool narrowing: the deny rules Crew seeds are keyed mcp__<server>__<tool>, so a same-named copy is narrowed like the projected one.

No per-backend branch at all, per harness-parity H6: the hedge is the weaker claim and needs no per-harness record, so nothing here reads a backend id or a spec field to choose a verdict.

One docstring in test_acp_mcp_ref_guard.py (the codex composition-path test) said the guard's definitive sentence is exactly right for a server the mirror withholds; it now says codex gets the qualified sentence, because the operator's own config.toml is the one thing the wire cannot rule out. That is a deliberate reversal of the earlier claim, on the codex record in agent_sdk/backends.py.

Review rounds (each row is that round's line, verbatim):

Round Line
1 Round 1 (advisory, all lanes clean on dd8acca627): reworded the leading clause (Opus 21d8fcaae5e2), carried the hedge into the doctor's no-mirror branch (Opus f44ebe40eb24), reworded the doc premise (Opus 0cc86c68a2e7), and made the hedge conditional on the spec's includeMcpJson (GPT d833401a6be9).
2 Round 2: r1's spec-conditional hedge reverted -- a backend-set complement is negative harness identity (GPT :145, Opus :233/:71) and includeMcpJson: false does not provably exclude the project mcp.json (Design); the hedge stays unconditional as FP settled it. The three wording fixes of Round 1 stay. Rebased after main's #14623 moved _doctor_unresolved_mcp_refs out of cli_doctor.py: the doctor hedge now lives in src/kiro_crew/doctor_checks/mcp.py (same sentence, same no-mirror branch, test follows it), and cli_doctor.py is no longer in the diff.
3 Round 3 (wording only, all lanes clean on 516d8048fe): corrected the contract's KAS sentence -- AcpRuntime._guard_unresolved_mcp_refs runs the same guard from _finish_create_session and load_session, so KAS is checked at runtime as well as statically, not "statically only" (GPT agent-host-contract.md:327); softened the guard's code comment on kiro-cli's global ~/.kiro/settings/mcp.json from "into every agent whatever the spec declares" to "by default (a spec may opt out of the global file with includeMcpJson: false)", as agent-spec-fields.md records it (Opus 21d8fcaae5e2, re-printed) -- comment text only, the emitted sentence and its unconditional shape unchanged; the contract's twin of that sentence is softened identically. Rebased onto main's #14651 (hermetic hop-guard cases on macOS), the heal for the inherited macOS (2) red.
4 Round 4 (wording only, the last text round; lanes clean on d2f15b149e, Opus advisory ×2): the r3 sweep is finished across the claim family -- actually about to receive, whatever the spec, into every agent, what the session … receive, no server this session gets, absent from the session -- over src/, test/, docs/ and website/docs; every occurrence describing this guard or its report field now carries the hedge wording (the two Opus items plus the sweep's own hits), the emitted sentence and its unconditional shape untouched (no condition, no set). The sweep ledger follows. The round lines moved into this table, verbatim, because What changed had passed the prepare-pr skill's 500-word prose cap (534) with Round 3; table rows are outside that count by the check's own rule.
5 Round 5: rebase onto main after the 16:15-16:29 merges; conflict in providers/mirrors/claude_code.py resolved main-wins; no behaviour change. Main gate at push -- main 6914028: Await Fast Gate failed on a cancelled Fast Gate run, no test red. Backend (3.12, 4) / (Windows) (4) red inherited from main: test_pid_reader_ratchet drift, tracked in #14776, not this PR's.
6 Round 6: rebase onto main e88e807786 to pick up #14778, the #14776 heal; no content change.

Round 4 sweep ledger:

Hit (line as the lanes cited it) Disposition
test/test_acp_mcp_ref_guard.py:367 docstring fixed -- kiro-cli loads the global file "by default (a spec may opt out with includeMcpJson: false)" (Opus)
test/test_acp_mcp_ref_guard.py:265, :702 docstrings fixed -- "still missing from Crew's projection"; "the array Crew sends a codex session"
src/kiro_crew/providers/mirrors/README.md:28 fixed -- refs compared against Crew's projection for the session; the line's hedged verdict spelled out (Opus)
src/kiro_crew/acp/mcp_session_report.py:418 (unresolved_refs field), :461 (record_unresolved_refs) fixed -- "named no server in Crew's projection"; the harness may still mount a same-named server, which the field does not read (Opus)
src/kiro_crew/agent_sdk/mcp_refs.py:23, :171 fixed -- "the server array Crew is about to send a session"; "no server in Crew's projection can satisfy"
src/kiro_crew/agent_sdk/backends.py:1466 (ACP_BACKENDS_SPEC_SERVERS_OFF_WIRE comment) fixed -- "what is not in it, Crew did not mount"; the harness may still mount one, so the line says "may", never "absent"
src/kiro_crew/acp/session_mcp.py:453 fixed -- "the array Crew sends a session"
docs/system-specs/modules/agent-host-contract.md:302 fixed -- "names nothing in Crew's projection for the session"
test/test_acp_mcp_ref_guard.py:374, :396 left alone -- the docstring states the hedge itself (the clause claims Crew's projection, not what the session receives) and the assert pins the old clause absent
src/kiro_crew/acp/mcp_ref_guard.py:47 left alone -- "whatever the spec put in the name" is about _log_safe redaction, not this claim
src/kiro_crew/doctor_checks/mcp.py:116, src/kiro_crew/providers/mirrors/README.md:291 left alone -- the no-channel row: pi-acp forwards no array, a registry-declared transport property, not this guard's verdict
src/kiro_crew/mcp_gateway/rewriter.py:2304 left alone -- "kiro-cli merges this file into every agent at runtime" is the rewriter's own reasoning about the global file, not this guard; the same absolute, outside this PR's scope
every other hit (KIROCREW_SESSION_KEY, steering and scratch-tree "into every agent"; "session receives" in unrelated subsystems) left alone -- describes something other than this guard or its report field

Backwards compatibility

Compatible: one log line changes wording (leading clause and closing sentence) on every backend, the doctor's no-mirror row changes wording the same way, and the comment and doc sentences that described the array change with it. Nothing parses those sentences; the session MCP report carries unresolved_refs as data and is unchanged. No vocabulary is added.

Tests

  • TestTheWarning::test_the_verdict_claims_only_what_the_wire_proves_on_every_backend, parametrized over ACP_BACKENDS_KNOWN, pins the full rendered line on every backend: the new leading clause, fields and remedy, the hedged sentence present, and neither "absent from the session", "nothing else to say so" nor "this session receives" anywhere. Red on main, green on this head.
  • TestUnresolvedMcpRefs::test_the_no_mirror_verdict_claims_only_what_the_wire_proves pins the hedge on the doctor's no-mirror branch for goose and kiro alike, with the broker-stub caveat and the registry pointer kept.
  • The redaction, alphabet-rebuild, bounded-length and call-site pairing tests in the guard file are unchanged and pass, so the sink hardening around the sentence is untouched.

Red-first evidence, on unmodified source with the wording test added (the claude case is shown; every backend rendered the same definitive sentence on main):

python -m pytest -n0 -q test/test_acp_mcp_ref_guard.py::TestTheWarning::test_on_claude_the_verdict_is_qualified_rather_than_asserted test/test_acp_mcp_ref_guard.py::TestTheWarning::test_every_other_backend_keeps_the_definitive_verdict_byte_for_byte
FAILED ...::test_on_claude_the_verdict_is_qualified_rather_than_asserted - assert 'may mount a same-named server from its own configuration' in "agent...
1 failed, 7 passed

On this head, -n0 one file at a time: test_acp_mcp_ref_guard.py 89 passed; test_cli_doctor.py 265 passed. Black ratchet (scripts/check_black_formatting.py), isort, flake8 and mypy on the changed Python files pass; HARNESS_BASE_REF=origin/main scripts/check_harness_parity.py clean; docs_lint.py --test and docs-lint.sh pass; push_guard.py --require-single-on-base reports a single commit on main.

Local review round on the first head (one lane at a time, read-only reviewers, CI's own contracts): GPT (gpt-5.6-sol) and Opus (claude-opus-5) each returned no blocking finding and one advisory on the claude-only form, that codex has the same own-configuration record and a branch per harness is the shape harness-parity H6 forbids; both are answered by the final shape, which has no per-backend branch. First Principles (claude-fable-5) ran four times as the shape narrowed: it moved the fix from a named set to the wire-fed set, deleted the "narrowing does not reach" clause (the deny rules Crew seeds are keyed mcp__<server>__<tool>, so a same-named copy is narrowed like the projected one), had every "entire/whole MCP surface" sentence in comments and docs reworded, and finally showed that kiro-cli's global settings-file record makes "absent" unprovable there too; this head is its Subtraction applied verbatim, so no fifth run was made. The codex composition-path test docstring, which said the definitive sentence was exactly right for a mirror-withheld server, is reworded to the hedge as a deliberate reversal on the codex record in agent_sdk/backends.py.

Manual verification

N/A — unit coverage sufficient: the change is one sentence emitted on every backend, and the tests render the log line through the real warn_unresolved_server_refs path for every known backend and the doctor row through _doctor_unresolved_mcp_refs.

Screenshots / video

no-visual-delta

Why no screenshot: backend log and kirocrew doctor wording only; no frontend file changed and nothing rendered differs.

Related Issues

Fixes #13652

Pattern harvest

Rule candidate: review-prompt
Pattern: a diagnostic asserts a negative ("absent", "nothing else to say so") about a surface its inputs do not observe; a log line may only claim what the data it read can prove, and when a harness can add to that surface behind Crew's back the weaker sentence is the one the evidence supports.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

Contribution License Agreement

@pepmach
pepmach requested a review from a team as a code owner September 28, 2026 03:57
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Sep 28, 2026
@github-actions

github-actions Bot commented Sep 28, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Fable 5) — ✅ PASS

Design-level review of efce193c8e6b93ff470fabfa7c2f1e3dba0ebabe — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Nothing to check.

[DESIGN-REVIEWED] efce193

@github-actions

github-actions Bot commented Sep 28, 2026 •

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of efce193c8e6b93ff470fabfa7c2f1e3dba0ebabe and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] efce193

False positive or not applicable? A repository writer can comment:
/ai-review override gpt efce193c8e6b93ff470fabfa7c2f1e3dba0ebabe: <one-sentence reason>

@github-actions

github-actions Bot commented Sep 28, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Fable 5) — ✅ PASS

Premise-level review of efce193c8e6b93ff470fabfa7c2f1e3dba0ebabe — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

First-Principles-Verdict: PASS

Only check left: no in-repo consumer parses the old sentence (grep found 0); confirm external alerting doesn't key on "absent from the session".

Not justified as shipped

  • Item 5 — rides along: the agent-host-contract KAS sentence ("checked at runtime as well as statically") is a separate doc correction, declared in round 3 and verified — AcpRuntime._guard_unresolved_mcp_refs exists at src/kiro_crew/acp/runtime.py:6953 with call sites at 7082 and 7655.

What this change ships

Inventory (7 items) — 6 justified

Intent: FIX — stop the unresolved-@server warning asserting harness-mounted servers are absent, a claim the base's own mcp_session_report.py comment ("NOT that the session has none") already contradicted.

  1. Warning's verdict hedges — "may still be served", never "absent" — justified
  2. Warning's leading clause claims Crew's projection, not what the session receives — justified
  3. Doctor's no-mirror row carries the identical hedge (same detector, static half) — justified
  4. Docs/comments sweep: "entire/whole MCP surface" → "only channel Crew has" (grepped entire MCP surface|whole MCP surface: 0 unfixed remain; survivors are the negated form) — justified
  5. Contract doc records KAS as runtime-checked too — rides along (declared, verified)
  6. Codex test docstring reverses its "sentence is exactly right, tools are absent" pin — justified (reversal declared, evidenced by the codex config.toml record in agent_sdk/backends.py)
  7. Full-line pin on every known backend plus doctor pin, so a per-backend branch cannot creep back (harness-parity H6, a documented invariant) — justified

Unconditional hedge on kiro-cli/KAS is derived, not symmetry: ~/.kiro/settings/mcp.json loads by default, so no backend's array/spec is provably the whole surface. doctor_checks/mcp.py:115's "every Crew tool is absent" is not a sibling — it claims only Crew's own channel, which Crew can prove.

[FIRST-PRINCIPLES-REVIEWED] efce193

@github-actions

github-actions Bot commented Sep 28, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5 Review — ✅ no blocking findings

Reviewed efce193c8e6b93ff470fabfa7c2f1e3dba0ebabe — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] efce193

Verdict parsed from the review's SHA-scoped output markers for commit efce193c8e6b93ff470fabfa7c2f1e3dba0ebabe.

False positive or not applicable? A repository writer can comment:
/ai-review override fable efce193c8e6b93ff470fabfa7c2f1e3dba0ebabe: <one-sentence reason>

@pepmach
pepmach force-pushed the fix/mcp-ref-guard-claude-harness-mounted-servers branch from dd8acca to 6c76dff Compare September 28, 2026 04:46
@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Sep 28, 2026
@pepmach
pepmach force-pushed the fix/mcp-ref-guard-claude-harness-mounted-servers branch from 6c76dff to 95549bb Compare September 28, 2026 05:42
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention merge conflict Branch has merge conflicts with its base — author must resolve before merge and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 28, 2026
@pepmach
pepmach force-pushed the fix/mcp-ref-guard-claude-harness-mounted-servers branch from 95549bb to 516d804 Compare September 28, 2026 07:41
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention merge conflict Branch has merge conflicts with its base — author must resolve before merge labels Sep 28, 2026
@pepmach
pepmach force-pushed the fix/mcp-ref-guard-claude-harness-mounted-servers branch from 516d804 to d2f15b1 Compare September 28, 2026 08:45
@github-actions github-actions Bot added readiness: passed Eligible automated validation passed for the current revision and removed readiness: checking Automated validation is still running labels Sep 28, 2026
@pepmach
pepmach force-pushed the fix/mcp-ref-guard-claude-harness-mounted-servers branch from d2f15b1 to 404dc50 Compare September 28, 2026 09:42
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision and removed readiness: passed Eligible automated validation passed for the current revision readiness: checking Automated validation is still running labels Sep 28, 2026
@pepmach

pepmach commented Sep 28, 2026

Copy link
Copy Markdown
Contributor Author

Accepted for acp/mcp_ref_guard.py:3 and the sibling docstrings: the module docstring still says a spec's refs "name nothing a session receives" — the claim this PR retired everywhere it describes the verdict ("names nothing in Crew's projection for the session"); round 4's sweep was a single-line grep over the claim family and the phrase wraps across a line break, so it was missed, as were the same-class summaries in client.py:6630/6634-35, runtime.py:6780 and mcp_refs.py:1. None of that text is emitted: the log line, the doctor row, the report field and every doc describing the verdict carry the hedge on this head, and the tests pin the old spellings absent from the emitted text. Reworded in #14682 as one docstring line each rather than by a fifth push that would restart the full round on a head green on every lane and CI. Kept, not deferred: doctor_checks/mcp.py:35 — "every Crew tool is absent while the harness works and nothing anywhere is red" narrates the three incidents recorded in providers/mirrors/README.md (what happened on those installs), not the guard's verdict about a session, so it is history, not the retired claim; and mcp_refs.py:18 describes the resolver's wire-only reading, which is Crew's projection claim itself.

iamwhatever
iamwhatever previously approved these changes Sep 28, 2026
@iamwhatever
iamwhatever enabled auto-merge (squash) September 28, 2026 16:23
@github-actions github-actions Bot added the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Sep 28, 2026
@pepmach
pepmach force-pushed the fix/mcp-ref-guard-claude-harness-mounted-servers branch from 404dc50 to 32440d0 Compare September 28, 2026 16:57
@pepmach

pepmach commented Sep 28, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto current main to clear a merge conflict in providers/mirrors/claude_code.py from today's batch. The change is otherwise identical to the approved head 404dc50d8a (rebase only, no behaviour change); the push dismissed the approval per the branch rules, so a re-approve is all that is needed.

@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: passed Eligible automated validation passed for the current revision merge conflict Branch has merge conflicts with its base — author must resolve before merge labels Sep 28, 2026
@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Sep 28, 2026
@pepmach

pepmach commented Sep 28, 2026

Copy link
Copy Markdown
Contributor Author
  • Item 5 — rides along: the contract doc's KAS sentence corrected ("KAS is checked at runtime as well as statically") span=3a82f566a686 — declared-and-kept

The correction removes a false sentence from the paragraph this PR already edits (the contract doc's hedge in docs/system-specs/modules/agent-host-contract.md), was declared in Round 3, and First Principles itself verifies it true against AcpRuntime._guard_unresolved_mcp_refs. Keeping a false statement adjacent to the PR's own wording would leave the doc contradicting the code it describes, so the correction stays in the change.

The unresolved-ref warning closed with a sentence the detector cannot
back: that the listed tools are absent from the session. The detector
judges the wire array (or, where the harness loads the spec itself,
the spec) alone, and no backend's array is provably the session's
whole MCP surface. Claude Code loads its own user- and project-scope
mcpServers and plugins beside the array Crew sends, codex-acp merges
Crew's array on top of what ~/.codex/config.toml declares, and
kiro-cli loads the global ~/.kiro/settings/mcp.json into every agent
whatever the spec declares -- so a same-named server may be serving a
listed ref while nothing of Crew's shows it.

The closing sentence now claims only what the wire proves, on every
backend: Crew's projection delivers none of those servers, and the
harness may mount a same-named server from its own configuration, so
a listed ref may still be served and the line cannot tell which. The
comments and docs that called the array the session's "entire" or
"whole MCP surface" say the same. The detector's satisfaction rule is
untouched, no harness configuration is read, and no backend id is
consulted.

Fixes #13652
@pepmach
pepmach force-pushed the fix/mcp-ref-guard-claude-harness-mounted-servers branch from 32440d0 to efce193 Compare September 28, 2026 18:45
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Sep 28, 2026
@pepmach

pepmach commented Sep 28, 2026

Copy link
Copy Markdown
Contributor Author
  • Item 5 — rides along: the contract doc's KAS sentence corrected ("KAS is checked at runtime as well as statically") span=de42ef2397c9 — declared-and-kept

The correction removes a false sentence from the paragraph this PR already edits (the contract doc's hedge in docs/system-specs/modules/agent-host-contract.md), was declared in Round 3, and First Principles itself verifies it true against AcpRuntime._guard_unresolved_mcp_refs. Keeping a false statement adjacent to the PR's own wording would leave the doc contradicting the code it describes, so the correction stays in the change.

@github-actions github-actions Bot added readiness: passed Eligible automated validation passed for the current revision and removed readiness: checking Automated validation is still running labels Sep 28, 2026

@bolichen97 bolichen97 left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewed on behalf of the maintainers (sweep 2026-09-28): readiness passed, required checks green on efce193, no blocking review threads, security posture unchanged. Approving and merging.

@bolichen97
bolichen97 merged commit 579582f into main Sep 28, 2026
97 checks passed
@bolichen97
bolichen97 deleted the fix/mcp-ref-guard-claude-harness-mounted-servers branch September 28, 2026 23:59
@github-actions github-actions Bot removed the readiness: passed Eligible automated validation passed for the current revision label Sep 29, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

mcp_ref_guard reports servers as absent on the claude backend when Claude Code loads them from ~/.claude.json

3 participants