agentleFS
HalemoGPAHalemoGPA/whatsapp-mcp-server/.claude/skills/verify-wire-claim/SKILL.md

verify-wire-claim

Skill9 starsChanged 4 months agoMITView on GitHub

Verify a claim about WhatsApp behaviour before writing it down. Use whenever about to state that a feature works, a flag is enforced, a message was view-once, a poll cap applies, or any assertion about what WhatsApp does. Also use when documenting a finding in docs/protocol-vs-app.md.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Test the send path: Call the tool. Capture the response.

    line 36
  2. Distinctive:

    Check the stored row: Read what actually landed in the message store, not what the tool said it wrote.

    line 44
  3. Very specific:

    Never upgrade a SOURCE to VERIFIED without an actual run.

    line 57

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Verifying a claim about WhatsApp

WhatsApp enforces most of what you see client-side. The app bets that what people see is reality, and for almost everyone that bet is correct. It is not correct here, because this server reads the wire.

A rendering is evidence of what a client chose to draw. Nothing more.

Never accept these as evidence

SourceWhy not
A screenshot, or what the desktop app showsDesktop and mobile enforce different things. A message once looked view-once on desktop for weeks while the inner flag was unset.
A comment in this repoThey rot. The two most confidently wrong statements in this codebase were comments that had been true when written.
A library's documentationIt describes the field's intent, not client behaviour. selectableOptionCount is documented as a maximum; no client enforces it above 1.
A tally or counter in the appFive phones once showed five different poll tallies while the wire said one number.
"The endpoint returned success"A poll vote returned Vote cast while the tally stayed at zero.

Accept these

  1. The actual protobuf, with the field number.
  2. A test you ran, with its output.
  3. An observation from a real phone, on the receive path.

The procedure

  1. Separate the three claims. "The field exists", "the server accepts it", and "the recipient's client honours it" are different, and a change can satisfy one and fail the others. Establish each on its own.
  2. Test the send path. Call the tool. Capture the response.
  3. Test the receive path on a real phone. This is the step people skip and it is where the bugs are. Sending through our own API exercises code that wraps and hashes locally and never decrypts or normalises. Two real bugs (all six poll-creation variants, and LID-vs-phone-number on votes) were found in minutes by a phone after the server-side tests passed clean.
  4. Check the stored row. Read what actually landed in the message store, not what the tool said it wrote.

Recording it

Findings go in docs/protocol-vs-app.md with two markers:

Confidence - VERIFIED (tested here, with the test), SOURCE (read in the protocol or library, not run), SPECULATIVE (inferred, not confirmed).

Status - EXPOSED (a tool uses it), BRIDGE_ONLY (the bridge can, no tool does), NOT_IMPLEMENTED (deliberately not built).

Never upgrade a SOURCE to VERIFIED without an actual run. If you are not sure which marker applies, it is SPECULATIVE.

The line on capabilities that deceive

Some of what the wire permits works by deceiving a recipient: forged quoted context, fabricated previews, spoofed attribution. That is different in kind from reading data already delivered to us, which is the whole point of this server. Both are technically available. Only one is the owner's own data.

Name which one you are doing. If it is the first, the answer is NOT_IMPLEMENTED, and the reasoning goes in the document rather than the field recipe.

When the user contradicts your model

They are probably right. "Desktop says only once, phone views it forever" was reported three times and explained away twice before it was recognised as the bug it was. The user can see a surface this server cannot.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.