[loopx-agent]

Goal And Delivered Outcome

• Goal/source and gap: The native progressive first read repeats the complete selected Todo note on a warm continuation, including large acceptance notes already delivered in the current context epoch.
• Observable before → after: The reader uses its existing observed-field reference for a verified, unchanged note. It delivers the note literally when it changes, a different Todo is selected, or the context epoch is lost.
• Issue/task and intended base: Standalone main slice related to C1: unify work sources and validate recovery #8266.

Author Declaration

• Written by: model_agent, OpenAI Codex (GPT-6 family).

Implemented against

• Specification and revision: Progressive Context Qualification at the tested base.

Criterion (spec clause)	Disposition	Symbol / path	Test or command
Reuse only observed source-bound material	implemented	reusableFirstReadFields and observedSectionDigest	Admitted File and SQLite continuations
Preserve the current task, authority metadata and original recovery capture	implemented	Native first-read planner and existing capture reader	Unrelated authority update, revised note, switched Todo and compaction recovery

• Self-check before submission: Verified the two canonical source shapes, exact note selection and byte-identical recovery from the pinned capture. Reviewed the protected public boundary and DCO.

Scope And Continuation

• This slice extends the existing native reader. Its digest binds the registry, runtime root, Goal, provider kind and selected Todo. The current provider revision still reaches the model.
• C1 can take this through main. Its broader admission and settlement integration and model-token and task-outcome evidence are tracked with C1: unify work sources and validate recovery #8266.

Validation

• Tested revision: fd37152339f0ebb317550e0049b8c56001e1d74e, paired against a3b589610a83f52024dd957c5071b0c257fab9b2 with the same fixture and CLI sequence.
• Run state: finished.
• Input classes: synthetic.

Check kind	Result	Public-safe evidence / limitation
real_entrypoint	passed	Real quota should-run, quota context-read, native hook events and successive durable writeback/spend on isolated fixture Goals.
real_backend	passed	File with the compact selected-source row and SQLite with the full canonical row. An unrelated Todo edit changes the provider revision before the warm read.
regression_parity	passed	File 42,631 → 21,888 bytes (48.7% less). SQLite 44,589 → 23,848 bytes (46.5% less). Both reduced the warm read from four pages to two. The base fails both new reuse assertions. The candidate reads the full changed note, identical note on a different Todo and the original note after compaction.
integration	passed	tests/control_plane/test_context_first_read.py passed all 80 tests. The three TypeScript context suites passed all 61 tests.
static	passed	TypeScript typecheck and public boundary scan passed. Standard diff-scoped premerge passed 13 selected smokes and its direct checks.

• Coverage and gaps: Byte and page figures cover reader stdout through EOF for the matched warm step. Model tokens, end-to-end latency and accepted task outcomes are unmeasured.

Frontend / Visual Evidence

• UI impact: none.
• Source data: none.

Type of Change

• Bug fix
• Test update

LoopX Area

• Control plane (goals, todos, quota, scheduler, registry, runtime)

Shared-authority RFC fixture impact

• N/A. This is a presentation-layer change.

Boundary Checklist

• The diff and PR materials use public-safe inputs and aggregate results.
• The validation uses isolated fixtures and does not duplicate maintainer-owned benchmarks.
• The change stays scoped to the existing reader.
• The RFC bilingual check passes.
• UI impact is marked none.
• The commit includes its DCO sign-off.