execution-ledger.md

Standards Agent Interface Execution Ledger

2026-09-05: Implementation Plan Created

Outcome: Planned direction recorded; implementation under this plan has not started.

Request: Improve focused navigation, explicit-fact discovery/explanations, and context-bound authoring without embeddings or an interpretation model.

Inspected evidence:

• The working-tree MCP transport exposes 19 generated native operations through AgentToolFacade; it does not yet implement the focused experience.
• StandardsEngine._route_value already emits base/rule/dependency causes and unresolved questions. Router reads with include_routing already expose fact definitions. Those are reuse opportunities, not new semantic authority.
• The existing read projection includes complete related rows, prerequisites, specialization, policy summary, content, and continuations.
• The contract README identifies one JSON shape authority and generated native and agent-tool projections.
• The Planning Workflow was read through the Engine's MCP dispatch path against an accepted snapshot. The plan follows its current lifecycle/artifact model.

Write set for this planning turn: This plan directory and its entry in docs/plans/README.md. Existing implementation changes and the unrelated proportionality-routing prototype are retained.

Verification: Local plan structure, milestone/acceptance coverage, and relative-link checks; no implementation acceptance claims are satisfied by writing this plan.

Commit: None created by this planning turn.

2026-09-05: Milestone 1 Accepted

Outcome: Focused navigation and compact exact policy reads accepted.

Evidence: Milestone 1. A1/A2 are satisfied; A7 has real-client navigation evidence and awaits authoring/recovery scenarios.

Scope: Included the previously implemented MCP transport and plan artifacts as prerequisites. The unrelated proportionality-routing prototype is excluded.

Deviation: Corrected a test oracle that equated independent snapshot IDs; production snapshot identity behavior was correct. No domain semantics changed.

Commit subject: feat(standards-engine): add focused agent navigation tools

2026-09-05: Milestone 2 Accepted

Outcome: Registered fact discovery and deterministic explanation projection accepted. A3 is satisfied; no Router or fact semantics changed.

Evidence: Milestone 2, 32 passing tests and an official MCP client walkthrough. Milestone 1 commit: e9bac84f.

Commit subject: feat(standards-engine): expose routing facts and explanations

2026-09-05: Milestone 3 Accepted

Outcome: One immutable context carries authoring state; Engine-derived continuations preserve explicit review, exact application, and separate recovery. All objective acceptance criteria are satisfied. No new context store or identity version was necessary.

Evidence: Milestone 3, including real-client normal publication, pending resolution, stale revision rejection, cold-process recovery, and supported Python verification. Milestone 2 commit: d97e1028.

Scope: Default MCP discovery now exposes 15 focused operations; all 32 remain available with --advanced. Skills use the focused surface. Personal client configuration and the unrelated prototype remain untouched.

Finding resolved: Terminal application outcomes must survive projection even when a later state read would fail. The wrapper preserves the native outcome directly, with a regression assertion and real publication/recovery coverage.

Commit subject: feat(standards-engine): bind authoring tools to immutable workflow context

Additional fixture correction: A broadened native review regression failed identically at Milestone 2 because it requested a nonexistent consumer disposition. The accepted fixture has only an audit-coverage obligation. Corrected the test to assert and resolve that obligation without changing production rules.

2026-09-05: Client Schema And Continuation Fixes

Request: Address hidden authoring schemas and unavailable continuation names before returning to the separate effectiveness plan.

Outcome: Inline nonrecursive input schema references from canonical contracts, preserving recursive validation. Engine-owned navigation projection uses focused tool names. Default instructions and unknown-outcome guidance use context-based recovery; advanced native operations retain readiness guidance.

Evidence: Client follow-ups, 37 focused regressions, 20 contract tests, Codex CLI 0.153.4 app-server schema consumption and same-snapshot route/read, and the 73-suite / 121-check Engine checkpoint. The local client check starts no model turn. This conversation needs a reconnect to inspect its final model-visible declarations.

Commit subject: fix(standards-engine): expose authoring schemas and focused continuations

2026-09-05: Preserve Nested Authoring Contracts In Tool Descriptions

Finding: The live reconnect proved the inline fix was partial: outer change-set fields appeared, but nested edits and evidence remained abbreviated as unknown. Focused guidance and navigation continuations were confirmed.

Change: Emit the exact canonical input closure in affected authoring tool descriptions, independently of client type rendering. No schema format, domain identity, authorization, validation, or persistence contract changes.

Evidence: Exact-description schema regression and actual configured Codex client preservation check; see follow-up evidence.

Commit subject: fix(standards-engine): retain nested input contracts in tool descriptions