See the boundaries.
Inspect JavaScript, web and extracted Electron artifacts. Keep extension contexts, candidate messages, source ranges and unresolved expressions distinct.
STATIC INSPECTION → SOURCE EVIDENCEAN INDEPENDENT AI REVERSE-ENGINEERING TOOLKIT
Understand
the mechanism.
Rebuild the behavior. Prove the result.
Connect Claude or Codex through local MCP. Inspect the artifact, challenge the explanation, and carry a testable behavior contract into a working implementation.
12 MCP tools · Agent skill · Direct CLI
THE QUESTION BEHIND THE SCREEN
“What survives when
the worker stops?”
01
Map the boundariesSeparate contexts. Follow candidate links.
02
Distinguish the explanationsChoose a probe that changes what you know.
03
Test the behavioral contractKeep the evidence. Expose the mismatch.
From a plausible explanation to a testable result
VISIBLE BOUNDARIESDECLARED ASSUMPTIONSREPRODUCIBLE CASESEDITABLE SOURCE
THE EXPERTISE, MADE INSPECTABLE
Follow the proof trail.
Explore recorded synthetic results from the toolkit’s own inspection, planning, verification and reduction operations.
RECORDED SYNTHETIC LAB
This explorer displays saved engine results. It does not run the target, attach to a browser or prove real service-worker termination. Check the release verification record for separately exercised runtime claims.
RECORDED OPERATION
Reading the lab…
Loading the supplied fixture results.
The recorded lab could not be loaded.
Reading recorded evidence.
No result loaded
Inspect the supplied data. Download the complete selected payload. Run the toolkit locally for your own case.
INTELLECTUAL WORK. EXECUTABLE TOOLS.
Go deeper than the visible path.
The toolkit gives an agent concrete machinery for an investigation, with an explicit boundary between static candidates and observed behavior.
Ask a sharper question.
Rank distinguishing probes against competing explanations, available capabilities and a declared budget. Missing predictions remain unknown.
HYPOTHESES → ADMISSIBLE PROBESKeep the mismatch visible.
Compare a finite behavior contract, preserve inapplicable evidence, and reduce a stable failing sequence without quietly changing the failure.
CONTRACT → REPRODUCIBLE CASEA NORMAL TASK. A MORE PRECISE INVESTIGATION.
Give your agent
the question that matters.
Install the skill, supply an authorized target and name the feature you want to understand or rebuild. The agent uses its available host tools and the toolkit’s explicit contracts.
Installation & first resultEXAMPLE TASK PROMPT
Use REA Pro Max on this local extension. Trace the popup, worker and storage boundaries, identify which observations would distinguish memory from persisted state, then rebuild the verified behavior in my project and leave a reproducible case for anything that still differs.
Use available runtime tools; retain unknowns when a required observation is unavailable.
A PRECISE SUPPORT BOUNDARY
Depth you can inspect.
Start with static JavaScript, Manifest V3 and extracted Electron inspection, plus the case, experiment, verification and reduction toolkit. Native and mobile analysis depend on separately supported providers and target-specific verification; a detected format is not a recovered program.
Hashes identify content. Reported observations keep their stated authority. Passing finite cases does not establish whole-program equivalence.
Read the release scope