reconbench

Case ExplorerCase 01

Reconstruct a browser-side Base64 workbench

best candidate reconstruction of Reconstruct a browser-side Base64 workbench by cc/opus
Fig. 00Development reconstructionby cc/opus, score 98.3

Evidence and provenance

Frozen evidence (3)

evidence capture: desktop-base64
E01 desktop-base64
evidence capture: desktop
E02 desktop
evidence capture: mobile
E03 mobile
  1. 01official source github.com
  2. 02live gchq.github.io
  3. 03reference-capture

Evaluation matrix

Score composition

Submission contract, install/build and policy are gates. Contract v2 preserves the full score vector instead of hiding functional checks, visual regions, responsive behavior, workflows and preservation in one value.

global 0.50 · regions 0.00 · functional 0.10 · responsive 0.10
workflow 0.30 · preservation 0.00 · floor 0.60

Active score weights are divided by their total. 1.00

Visual checkpoints (2)

reference capture for desktop
Fig. 01desktop1440×900 · viewport · motion reduce
reference capture for mobile
Fig. 02mobile390×844 · full page · motion reduce

Contract v2 evidence

1 workflow · 4 ordered steps · 0 weighted regions

EvidenceContractWeight
base64-recipeinitial-empty assert ×0.8 → add-operation drag ×1.5 → enter-input fill ×2.0 → base64-result screenshot ×0.32.0

Deterministic checks (5, total weight 3.0)

CheckKindDimensionTargetWeight
operation-presentselectorfunctionaldesktop0.5
recipe-presentselectorfunctionaldesktop0.5
input-presentselectorfunctionalmobile0.5
output-presentselectorfunctionalmobile0.5
mobile-no-overflowno_horizontal_overflowresponsivemobile1.0
Development diagnostics · 10 recorded attempts
10 / 10completed attempts
0infrastructure failures
0timeouts
20.7–98.3observed overall range
case idmodelmodel requestedmodel resolution statusruntimestatus
Pin to compareDateAgentVariantStatusVisualFunc.OverallTimeCost
2026-08-30claude_code/claude-opus-5 · high—completed98.7100.098.336m 22s$12.80
2026-08-30claude_code/claude-sonnet-5 · high—completed64.6100.055.87m 22s$1.79
2026-08-30codex/unverified · high—completed91.4100.089.26m 34s—
2026-08-30codex/unverified · high—completed62.0100.022.57m 07s—
2026-08-30codex/unverified · high—completed92.7100.090.99m 09s—
2026-08-30claude_code/claude-opus-5 · high—completed97.7100.097.127m 58s$10.78
2026-08-30claude_code/claude-sonnet-5 · high—completed71.6100.064.58m 34s$2.54
2026-08-30codex/unverified · high—completed60.6100.020.77m 26s—
2026-08-30codex/unverified · high—completed84.2100.080.23m 53s—
2026-08-30codex/unverified · high—completed93.4100.091.89m 06s—
Candidate protocol preview
Access-restricted Pilot

The rendered Candidate prompt is published for inspection. The downloadable starter workspace, injected Skill files and executable Harness remain restricted, so this page does not claim that an external reader can reproduce a graded run.

Skills declared design-case-implementer

Rendered Candidate prompt

Runtime-agnostic prompt record.

Use the ReconBench candidate skills `design-case-implementer` in that order. Read those skill files if your workspace provides them, and otherwise follow the same process with your own tools.

You are the candidate in ReconBench task `cyberchef-base64-workbench` (Reconstruct a browser-side Base64 workbench).

Read `task/brief.md` and every other file under `task/` before changing code.
Implement the result only under `submission/`. The starter may be replaced when the task allows it.
Do not inspect parent directories, benchmark harness files, grader code, reference images, or any oracle material.
Do not call write endpoints belonging to a reference site. Keep forms, counters, and similar behavior local.

Dependencies install offline from a vendored cache that matches the starter lockfile.
Do not add dependencies; the install will fail and the network is not yours to use.

`tools/screenshot.mjs` renders your work in the browser build the evaluator uses:

    node tools/screenshot.mjs --url http://127.0.0.1:4173/ --out shots/desktop.png \
      --width 1440 --height 900 --full-page

Before finishing:
1. Run the requested build command.
2. Start the app and capture every target viewport in the task with that tool, then read the captures back.
3. Exercise the required interactions and inspect the console errors and blocked requests it reports.
4. Write `submission/submission.json` using the contract in the implementer skill.

The harness will evaluate the filesystem after this process exits. Do not leave a dev server running.

Not publicly reproducible in this Pilot

Starter workspace
access restricted
Injected Skills
access restricted
Harness command
withheld until a runnable public bundle is released
Sealed evaluator
never Candidate-visible
Appendix ATask brief — exactly what the candidate agent receives

CyberChef Base64 workbench

Reconstruct the frozen browser-side transformation workbench shown in the supplied evidence. This is an operational desktop tool, not a marketing page. Match the four-area desktop layout at 1440x900 and the full-page compact phone layout at 390x844. On phone, Operations and Recipe share the top row while the Input and Output work areas continue below them without horizontal overflow.

The visible areas are:

  • the Operations library with favourites, categories, counts and search;
  • the Recipe workspace with its action rail and Auto Bake control;
  • the Input editor and status controls;
  • the Output viewer and status controls.

Implement this observable workflow:

  1. Drag To Base64 from Operations into the Recipe workspace.
  2. Enter ReconBench second case in Input.
  3. With Auto Bake enabled, Output becomes UmVjb25CZW5jaCBzZWNvbmQgY2FzZQ== and matches evidence/desktop-base64.png.

Expose To Base64 operation, Recipe workspace, Input data, and Output data as accessible names on their corresponding interactive or content surfaces. Make the operation draggable and the recipe a working drop target. The input must be a directly editable textarea or contenteditable element. The output must be a non-form rendered text surface such as div, pre, or a non-editable contenteditable; do not put Output data on input or textarea. The evaluator reads the same visible text a user reads rather than a hidden form value.

Keep every behavior local. Do not embed CyberChef, copy its source, use a screenshot as the implementation, call production APIs, or load runtime assets from the network. Do not implement unrelated cryptographic operations.

Run npm run build, then write submission/submission.json according to the supplied contract.