[loopx-agent]

Goal And Delivered Outcome

Opening cited command evidence currently shows an escaped JSONL event, making the printed result hard to inspect. The shared Research citation reader now presents the exact output and exit code first, with command arguments and the complete original record available in place.

This completes a bounded part of progressive research reading: ordinary report or deep explanation → citation → readable evidence → return to the same citation. Target: main at 8a718ca5f17b8036a57c6d4e88a02185eaf4581d.

Author Declaration

• Written by: model_agent — OpenAI Codex, GPT-6 family.
• Specification: docs/development/design.md (Research Workspace, Explanations On Demand and Populated Page Review) and docs/development/frontend-delivery.md (Task-first delivery), at 8a718ca5f17b8036a57c6d4e88a02185eaf4581d.

Criterion	Disposition	Consumer	Evidence
Read complete cited content and preserve its origin	implemented	ResearchCitationPanel in goal-research-analysis.tsx	Real pack registration, source verifier and HTTP citation readback; exact output and raw text assertions
Readable requested detail, with supporting protocol data on demand	implemented	Local CitationContent and existing source styles	16px output, desktop/390px captures, long output uses page scrolling
Keep original context and failure states	implemented	Existing citation controller	Keyboard return, negative/empty output, literal fallback and full research-history regression

Scope And Continuation

The presenter recognizes event_msg / item_completed / CommandExecution records. Unknown, mixed, malformed or numerically unsafe records retain the complete literal fallback. Source eligibility, hashes, line ranges and analysis execution remain with the existing backend owner.

Complete within this scope. Broader analysis quality and release acceptance remain separate. This is a local renderer in the existing component, with no new API or generic event-decoding module; later formats can retain the fallback until explicitly supported.

Validation

• Tested source: b398a068f640d4e13d93fdad9c134c7093f271f7; the packaged tests and captures used these same committed component/CSS/fixture contents, verified against source after commit.
• Run state: finished. Input classes: synthetic, authorized_private_read_only.

Check kind	Result	Evidence / limitation
static	passed	Typechecked packaged build, bundle source verification and CSS custom-property checks
real_backend / integration	passed	Packaged research-native-reading: actual registration, qualified source and HTTP owner; no intercepted citation responses
regression_parity	passed	Plain evidence remains literal. Known output, raw equality, failure/empty/long content, mixed/malformed/unsafe fallback, script escaping, Chinese/English and exact focus return covered
real_entrypoint	passed	Complete packaged research-history scenario, including existing read failure/recovery paths
manual	passed	Same synthetic source before/after at 1280×900 and 390×844; pixels and PNG metadata inspected. Authorized read-only material confirmed unchanged literal output and return context; private evidence is not attached
static / manual	blocked	Standard premerge: all four direct checks, seven catalog checks and public text boundary passed; four binary-image review holds remain for the exact-head reviewer

The new presentation changes no analysis or experiment results. Installed App adoption, reader task-time benefit and scientific interpretation quality are unqualified. No merge or auto-merge is requested.

Frontend / Visual Evidence

• UI impact: changed. Source data: synthetic.
• Packaged reader first viewport; before assets frozen at 8a718ca5f17b8036a57c6d4e88a02185eaf4581d, after source b398a068f640d4e13d93fdad9c134c7093f271f7. No public hosted preview is available.
• Output carries the reading hierarchy. Per-record exit codes retain failures; arguments and raw envelopes move behind their own disclosures. Complete content remains selectable, with no inner height cap on the primary output. Existing loading/error and close/return behavior is retained.
• Craft review: one source-reading region, existing type/color tokens and hairlines, no new decorative container. Long populated output and negative/unknown states were inspected on desktop and narrow screens. Goal Overview composition is unchanged; no broader showcase or installed-App claim.

View	Before	After
Desktop
Narrow

Type of Change

• Bug fix
• Test update

LoopX Area

• Public docs or presentation surface

Technical Direction

Operator surface; the existing Research evidence-reading contract. Shared-authority RFC fixture impact: N/A; no authority-store, routing, compatibility or control-plane decision change.

Boundary Checklist

• Public-safe diff, description and synthetic screenshots; no private state, raw research traces, credentials or local paths.
• No duplicated benchmark work; existing analysis/source owner reused.
• Scoped to the source-reading defect.
• No paired RFC changed; bilingual diff check passed.
• Visual evidence included above.
• Every commit has a DCO sign-off.