@tea-agent/loop-agent 0.42.0 → 0.43.0-next.2

package/skills/frontend-contract/references/contract-protocol.md

@@ -8,8 +8,8 @@
   source-fidelity ledger. Do not infer fragment IDs from prose.
 - Record observable outcomes, applicable UI states, runtime/environment constraints,
   non-goals, risks, and verification expectations without choosing implementation.
-- Classify each requirement as `explicit`, `repository-resolvable`, `assumption`, or
-  `blocking`. Blocking facts name their owner and evidence.
+- Runtime owns requirement text, source binding and disposition. Use typed questions
+  and constraints for uncertainties; blocked terminal facts name their owner.
 
 ## Evidence and handoff
 
@@ -29,6 +29,22 @@ enumerate or downgrade them.
 
 ## Terminal
 
-Commit facts in small batches and call `finalize_contract` exactly once with
+Commit facts from the complete current scope, then `complete_contract_scope` only
+after its evidence, constraints and source-bound deliverables are recorded. Scope
+checkpoints survive interruption; confirming an ID alone is not completion.
+Call `finalize_contract` with
 `ready`, `ready-with-assumptions`, or `blocked`. If the task is too large for one
 bounded writer, record a split proposal instead of broadening the contract.
+
+A successful receipt means the fact was saved durably. Replays are idempotent;
+changed facts require explicit `replace:true`. Correct rejected calls and retry,
+but do not mutate after a successful terminal. Never drop tail conditions to fit.
+
+## Execution ownership
+
+`record_requirement` may include `execution:{groupId,kind,summary}`. Use `behavior`,
+`constraint`, or `exclusion`; reuse an existing group only when semantics agree.
+Keep permission, threshold and failure-path differences separate. Do not merge
+source IDs or independent acceptance conditions. Unclassified requirements remain
+independent. Later scopes receive the shared group index. A group is a planning
+proposal; independent design review must check every member's complete text.

package/skills/frontend-design-review/SKILL.md

@@ -9,26 +9,27 @@ references:
 
 # Frontend Design Review
 
-For frontend design review nodes. Read the checklist and the non-blocking contract
-capacity diagnostic first. Audit either the canonical contract or, when the
-diagnostic recommends indexed reading, its hash-bound field sections. Check the
-bound Mock strategy, task bounds, and traceable design evidence supplied by the
-upstream context. Do not rediscover task sources, search specification roots, or
-replace frozen evidence with nearby-code conventions. Use only exact, bound evidence
-paths when additional reading is necessary; never invent connector results.
+Audit the canonical contract or its hash-bound indexed sections, guided by the
+capacity diagnostic. Check Mock strategy, task bounds and exact bound evidence.
+Do not rediscover sources, search specification roots or invent connector results.
 
 ## Authoritative Terminal
 
-Complete the review with exactly one committed typed tool call:
+Submit findings one at a time with `record_design_finding` using stable IDs.
+Use `replace:true` for revisions. The terminal automatically includes saved findings;
+do not repeat their array. Correct rejected calls, then commit one successful terminal:
 
 - `approve_design` when no blocking design defect remains;
 - `request_design_changes` when a blocker exists. It must include a typed
-  `issueCategory`, at least one `evidenceRef`, and non-empty findings.
+  `issueCategory`, at least one `evidenceRef`, and non-empty findings (saved or supplied).
 
 The typed terminal is the only authoritative outcome. Do not require or emit a
 `VERDICT:` line, JSON verdict, revision patch, or a second final-design-review
-verdict. A requested design change ends this run; the recovery controller creates
-any bounded follow-up run from the appropriate phase.
+verdict. Requested changes end this run; recovery owns the bounded follow-up.
+
+Check every execution group against its members' complete outcomes. Permissions,
+thresholds and error paths remain independent ACs; a shared VT must prove them all.
+Constraint/exclusion labels never waive source-required behavior evidence.
 
 ## Blocking Conditions
 

package/skills/frontend-plan/SKILL.md

@@ -4,7 +4,7 @@ description: Use only for the tool-only frontend decision planning node.
 references:
   - path: references/decision-contract.md
     required: true
-    maxChars: 5600
+    maxChars: 6800
   - path: references/design-decisions.md
     required: true
     maxChars: 3400
@@ -35,5 +35,5 @@ here with the same id and `replace=true` (union backfill) before finalize. A
 `duplicate` receipt is a correction invitation, not a prohibition.
 
 The runtime owns protected skeleton fields, path/command validation, materialization,
-and final schema checks. Finish with exactly one `finalize_plan`; return no Markdown
+and final schema checks. Finish with one successful `finalize_plan` (correct rejected facts and retry); return no Markdown
 narrative. Record a genuine evidence gap when frozen facts are insufficient.

package/skills/frontend-plan/references/decision-contract.md

@@ -95,9 +95,11 @@ facts".
 
 ## UX vocabulary (registry first)
 
-- Record `record_state_registry` BEFORE `record_state_flow`: one global vocabulary.
-  Coverage may slice by requirement; UX must not. Use stable kebab-case behavior
-  names, never AC-number-derived names or renamed committed concepts.
+- Record `record_state_registry` BEFORE any `record_state_flow`: one compact
+   global vocabulary of UI-state and interaction names. Details use complete execution
+   group scopes; reuse shared names across scopes, never one name per AC or a
+   rename of an already-recorded concept (kebab-case behavior-domain names, e.g.
+   `planner-task-edit`, `focus-queue-move`).
 - UI states use the contract's declared authoritative ids (`declaredUiStates`
   in the plan input) when the source declares a state table.
 - Retry slices see `committedUx` in the plan input: reuse those exact names.
@@ -123,11 +125,22 @@ widen the write set.
 
 The runtime owns `schemaVersion`, `sourceBinding`, risk, canonical target files,
 `requiredDeliverables`, aliases, command allowlisting, path containment, and final
-validation. Submit only editable decision facts, then call `finalize_plan` exactly
-once. The deterministic design policy materializes the canonical implementation
+validation. Submit editable decision facts, then call `finalize_plan`; correct
+rejected facts and retry until one successful terminal commit. The deterministic design policy materializes the canonical implementation
 contract: every file in the Contract's explicit source-bound delivery inventory
 must be planned (`requiredDeliverables` gate). Permissions, prohibitions, examples
 and read-only references do not create delivery obligations. Plan receives this
 inventory as `requiredDeliverables`; it cannot add, remove or reinterpret it.
 Missing files and UX name collisions are reported by finalize_plan before commit
 and checked again by the policy shell.
+
+
+Use `record_plan_group_coverage` for a complete declared execution group with shared
+targets/VTs; runtime expands to all canonical members. Submit individual coverage
+for differences. Constraints/exclusions create no invented UI or tests; keep genuine
+evidence gaps, and never weaken behavior evidence required by the source.
+
+Submit Mock policy with `record_mock_api` and empty endpoints, then one endpoint per
+`record_mock_endpoint`. Use `replace:true` for revisions, plus `remove:true` to
+withdraw an endpoint. Use stable finding/fact identities; successful receipts mean
+durable commits. Resume missing work instead of regenerating full JSON or history.

package/skills/frontend-review/SKILL.md

@@ -9,25 +9,24 @@ references:
 
 # Frontend Review
 
-Use for `frontend-review-pi`; read the findings guide and capacity diagnostic first.
-Inputs are `contracts/frontend-review-context.json`, its hash-bound contract
-reference/index and diff fragments, and only source files named by that diff. Use
-indexed contract sections when recommended. Do not reopen task sources, plans,
-design-review prose, OpenSpec, or `ai_workspace/`; the bound context is authoritative.
+Use `contracts/frontend-review-context.json`, its hash-bound contract index and diff,
+and only files named by the diff. Read the findings guide and capacity diagnostic;
+use indexed sections when advised. Do not reopen task sources, plans, OpenSpec,
+design-review prose or `ai_workspace/`.
 
 ## Authoritative Terminal
 
-Complete the review with exactly one committed typed tool call:
+Submit findings one at a time with `record_review_finding` using stable IDs.
+Use `replace:true` for revisions. The terminal automatically includes saved findings;
+do not repeat their array. Correct rejected calls, then commit one successful terminal:
 
 - `approve_review` when no Critical/Important finding remains;
 - `request_review_changes` for a defect, missing required evidence, forbidden
   write, or unmet acceptance criterion. It must include a typed `issueCategory`,
-  at least one `evidenceRef`, and non-empty findings.
+  at least one `evidenceRef`, and non-empty findings (saved or supplied).
 
-The typed terminal is the only authoritative outcome. Do not require or emit a
-`VERDICT:` line or response-text JSON verdict. A request for review changes ends
-this run and is handled by the recovery controller rather than a same-run repair
-branch.
+Only the typed terminal controls the outcome; no `VERDICT:` line or JSON verdict.
+Requested changes end this run; the recovery controller owns follow-up.
 
 ## Review Scope
 

package/skills/frontend-scout/references/scout-evidence.md

@@ -21,3 +21,7 @@ the exact evidence. Do not recursively inspect unrelated directories.
 Commit `complete` only when the target surface is sufficiently proven for Plan to
 reuse without repository discovery. Otherwise commit `blocked` and list every
 unresolved path or ownership question. Do not shift discovery work into Plan.
+
+A successful typed receipt is durable. Reuse only facts bound to the same source
+and fresh file evidence; changed or missing evidence requires Scout refresh. Do
+not replay stale success or repeat complete discovery merely because a reply was lost.