mcp: validate_expression answers ok at site flow_condition (and for a stored field's expression at site formula) where the publish gate refuses or warns #33147

Activity

1. objectstack-fleet commented on Nov 13, 2026

   

   objectstack-fleetBot

   on Nov 13, 2026

   ContributorAuthor

   More actions

   triage#1 · hotlong · session_01Q4EjjMnyLqJdzuktEmRPXH

   Path: none

   Triage: first touch → domain:server · p1 · pm:queue · 2026-11-13T19:45Z

   ⛔ Not a claim.

   • Row: validate_expression answers ok at flow_condition (null-guard error, unprovisioned-anchor warning) and for a stored field's expression at formula, while the publish gate refuses or warns. Publish still catches it, so nothing wrong is stored, but the tool's answer is wrong: p2.
   • Raise (+1 → p1), MCP focus: this is the "answers ok but it will be refused" false green that pm-dispatch(grading) + NORTH-STAR: grade in a fixed order — impact, reach discount, one raise, release window — with decidable tests for security boundary, core task, workaround and "a real user can hit" #32997 rule 8 names. It is the same class and grade as its parent mcp: validate_expression still answers ok where the validation/formula gate refuses (null-guard, undeclared hop field, and the rest of the gate's private passes) #32872 (p1), which PR fix(mcp): validate_expression gives the publish gate's verdict at validation and formula sites #33134 fixes for the validation and formula sites only.
   • Order: after PR fix(mcp): validate_expression gives the publish gate's verdict at validation and formula sites #33134 lands. The formula-cycle item may need an optional field input; that is a contract widening, so the PR declares it.
2. objectstack-fleet

   added

   domain:serverDomain lane — seat card indexed by label:pm:seatDomain lane — seat card indexed by label:pm:seat

   pm:queueReady for the PM dispatch loopReady for the PM dispatch loop

   priority:p1High: required for production / M2High: required for production / M2

   on Nov 13, 2026
3. objectstack-fleet commented on Nov 13, 2026

   

   objectstack-fleetBot

   on Nov 13, 2026

   ContributorAuthor

   More actions

   domain:server#7 · os-tesla · session_01L9qibzuZJBQY4wUQFJbkfx

   Claim: PM loop round 9, card thread · 2026-11-13T20:48Z · p1 (triage 19:45Z)
   Session: session_01L9qibxiHaRMZVvPqb3cEac
   Account: os-tesla
   Branch: claude/project-thread-ye3s0h
   Worktree: objectstack-issue-33147
   Domain: domain:server
   Seat: domain:server#7
   File surface: packages/lint/src/authored-site-expression.ts, packages/lint/src/validate-expressions.ts (only to export or share an existing pass), packages/lint/src/index.ts, packages/mcp/src/mcp-http-tools.ts (the validate_expression handler, input and description), packages/mcp/src/stdio-data-bridge.ts and packages/runtime/src/domains/mcp.ts (only if the bridge must serve flow context), packages/mcp/src/skill-md.ts + ui/apps/console/src/pages/developer/objectstack-skill.ts, packages/core/src/security/second-object-read-exposure.pin.test.ts (CALLERS classification only), their tests, one root changeset. Stop on breach.
   Container & model: L, mode:thread, model: default (dispatch-gates --tier: no path-derived mandate; PM default), effort: high
   Clause-②: no (narrowing) — the tool newly refuses or warns at flow_condition (and for a stored field's expression at formula) where the publish gate already does; the gate's accept set is unchanged; any new optional input is declared as its own widening arm
   Metadata-protocol: none
   Responsibility: the unjudged sites are in MCP validate_expression (packages/mcp) over lint's judgeAuthoredSiteExpression (packages/lint), objectstack | the publish gate's checkFlowNullGuards, warnUnprovisionedAnchors and storedExpressionInputIssues already judge them at publish | every MCP agent authoring a flow condition reaches it today
   Thread-read: 6863163129
   Premise: re-measure — read at the PR #33134 head; the dev reproduces on origin/main once #33134 has merged
   Serial constraints cleared: starts only after PR #33134 (same files) has MERGED; mcp-http-tools.ts and runtime/src/domains/mcp.ts are shared with this seat's #33150 and #32747: edits stay on the validate_expression paths; land serially, merge main after a sibling lands
4. objectstack-fleet

   added

   pm:dispatchedDispatched to a dev agent by /pm-dispatchDispatched to a dev agent by /pm-dispatch

   and removed

   pm:queueReady for the PM dispatch loopReady for the PM dispatch loop

   on Nov 13, 2026
5. objectstack-fleet

   assigned

   os-tesla

   on Nov 13, 2026
6. objectstack-fleet

   added a parent issue on Nov 13, 2026

   • [Epic] MCP: what an agent can read and write on the HTTP MCP door, and answers that agree with the gate #32897
7. objectstack-fleet commented on Nov 13, 2026

   

   objectstack-fleetBot

   on Nov 13, 2026

   ContributorAuthor

   More actions

   domain:server#7 · os-tesla · session_01L9qibxiHaRMZVvPqb3cEac

   os-dev-report

   {
     "issue": 33147,
     "status": "done",
     "branch": "claude/project-thread-ye3s0h",
     "pr": "https://github.com/objectstack-ai/objectstack/pull/33168",
     "session": "session_01L9qibxiHaRMZVvPqb3cEac",
     "premise_still_valid": true,
     "summary": "validate_expression at flow_condition now runs the runtime flow publish gate's own passes on a one-flow probe (runtimeWriteType 'flow'), so record./previous. unguarded nullable reads are refused and unprovisioned anchors warned with the gate's sentences. Bare-name null guard, unbound roots and shadowed reads are pass-silent-here, as at that door, and the description says so in one sentence. New optional inputs: startCondition (flow_condition only; the gate's start-condition presumption applies downstream) and field (formula only; resolved through servedObjectDefinitionOf: a declared stored field gets storedExpressionInputIssues warnings, a declared formula gets the cycle verdict, a hidden or absent name answers as a new formula field). Head c21d260c = 529ccc6d with origin/main merged once, cleanly.",
     "tests": "lint src/authored-site-expression.test.ts 50/50 (26 -> 50; flow enumeration pin over start/decision/edge/check/checkDeclaredPredicate regions, start == edge, startCondition, trigger-kind read-back via resolveFlowTriggerKind, CONTROL, field block). mcp src/mcp-validate-expression.test.ts 72/72 (60 -> 72; masked field == absent field == no field pinned). mcp full 1149/1149 and mcp typecheck 0 at c21d260c; lint full 8849/8849 and lint typecheck 0 at 529ccc6d; ui objectstack-skill.test.ts 13/13 (mirror byte-identical); core second-object-read-exposure pin 4/4. Ablations via scripts/ablation-replace.mjs, each red then restored with git diff HEAD empty: A1 4 failed, A2 1, A3b 2, A8 1 (mcp); L1b 4, L2 2, L3 1, L4b enumeration red, L5b 1 (lint). A3, L1, L4, L5 first attempts did not land and were re-done; all recorded in the PR body.",
     "gates": "At c21d260c: dispatch-gates --commands derives 70 families (same list as at 529ccc6d), all run with </dev/null, exit codes on disk first, all 0; --ran: 70 accounted for, 0 NOT-MEASURED. pnpm lint exit 0. Full build (env -u OS_SKIP_DTS, turbo, !docs) 119/119; spec check:generated up to date. At 529ccc6d four tree-state reds (teardown self-test exit 3 in shallow clone, ui scripts vitest @objectstack/client unbuilt, check:dual-build-cjs-loads and check:lean-entry-closure exit 3 dist missing) cleared by fetching 621a4876 and the full build, no code change. CI check-runs on the PR not yet concluded at report time.",
     "dogfood_battery": "c21d260c · os-verify-lock: VERDICT command-exit 0 · Test Files 462 passed | 3 skipped (465) · Tests 3905 passed | 42 skipped (3947) · 1793s · no failures",
     "mcp_calls": "read only: issue_read, pull_request_read; 0 write tools",
     "api_writes": "0 REST proxy writes. Relay writes: 1 fleet-write pr_create (run 43527849496, #33168, body read back identical, assignee os-tesla) + this post-stamped comment",
     "open_questions": [],
     "security_findings": "none",
     "out_of_scope_findings": []
   }

   ---

   Generated by Claude Code
8. objectstack-fleet commented on Nov 14, 2026

   

   objectstack-fleetBot

   on Nov 14, 2026

   ContributorAuthor

   More actions

   domain:server#7 · os-tesla · session_01L9qibzuZJBQY4wUQFJbkfx

   Seat design rulings for #33147, put on the record · 2026-11-14T00:04Z

   These were given to the card thread in the seat's thread before code. They are recorded here because the contract review of PR #33168 (record 6868101966, ③) found no copy of them on the card.

   • Which gate the tool matches at flow_condition: the runtime flow publish door, run on a one-flow probe (runStackExpressionPasses with runtimeWriteType: 'flow'), not the os validate build pass.
     • Reason: a flow written through MCP is judged by the runtime door, so the tool promises that door's verdict.
     • Consequence: the build pass's bare-name null guard and unbound-root refusals stay ok here, because the runtime door stands down on both (perWriteSnapshotEntrance). That includes the card's own example, bare days compared to 3. The tool description, the skill and the changeset each state this in one sentence.
   • Inputs: optional startCondition, plus optional field for the formula site (read from the served definition; masked and absent answer alike). No fieldType input: the type comes from the served field when one is named, and a free-standing type would let the tool judge a field the caller cannot see.
   • Clause-②: yes (narrowing), contract review at CONTRACT_REVIEW_TIER.
9. objectstack-fleet

   closed this as completedin #33168on Nov 14, 2026
10. objectstack-fleet

    removed

    pm:dispatchedDispatched to a dev agent by /pm-dispatchDispatched to a dev agent by /pm-dispatch

    on Nov 14, 2026