github/awesome-copilot / skills/webmcpify
webmcpify
VerifiedWebMCP agent skill for curated core coverage or route-by-route parity — inventory an existing web app, integrate approved tools, then inspect, verify and heal them in a real browser. Use for "webmcpify", "add WebMCP", or "expose app actions to AI agents".
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 31.3 KB |
| references/discovery.md | markdown | 8.5 KB |
| references/heal.md | markdown | 8.3 KB |
| references/integrate.md | markdown | 10.7 KB |
| references/inventory.md | markdown | 9.9 KB |
| references/native-compatibility.md | markdown | 5.3 KB |
| references/reverify.md | markdown | 11.3 KB |
| references/runtime.md | markdown | 8.7 KB |
| references/security.md | markdown | 5.0 KB |
| references/verify.md | markdown | 15.4 KB |
| references/workbench.md | markdown | 5.0 KB |
| scripts/workbench.mjs | script | 5.6 KB |
| templates/mutation-journal.js | script | 22.3 KB |
| templates/mutation-journal.ts | script | 25.8 KB |
| templates/webmcp-compat.js | script | 2.5 KB |
| templates/webmcp-jsx.d.ts | script | 2.3 KB |
| templates/webmcp-workbench.js | script | 37.9 KB |
| templates/webmcp.d.ts | script | 6.1 KB |
| templates/webmcp.spec.ts | script | 18.9 KB |
| templates/webmcpify.js | script | 12.0 KB |
| templates/webmcpify.ts | script | 12.5 KB |
| templates/well-known-webmcp.json | config | 1.2 KB |
| tests/README.md | markdown | 1.3 KB |
| tests/workbench-cli.test.mjs | script | 4.5 KB |
| tests/workbench.test.mjs | script | 5.4 KB |
references/native-compatibility.md
View on GitHub# Native compatibility — documentation checked 2026-10-12
WebMCP is evolving. These are separate evidence levels; simulation results are
not native browser acceptance.
## Current primary documentation
The [CG draft](https://webmachinelearning.github.io/webmcp/) specifies callback
options containing an execution signal, optional input defaulting to `{}`, and
JSON serialization of successful imperative results. A callback returning `"done"`
therefore produces the JSON string text `"done"` at `executeTool`, including the
quotes; a callback object produces serialized JSON. Callback values and native
client results are different boundaries. Browser-owned `null` for declarative
navigation is separate from the runtime guard against absent imperative results.
The [Chrome imperative guide](https://developer.chrome.com/docs/ai/webmcp/imperative-api)
(updated 2026-09-21) documents execution cancellation via callback `{ signal }`,
Chrome 153 preserving in-flight executions when tools are unregistered, Chrome 155
deprecating JSON-string input in favor of object input, and the `debugging`
annotation from Chrome 156. These version claims are documentation, not proof
from the installed Chrome 150 run.
The [Chrome declarative guide](https://developer.chrome.com/docs/ai/webmcp/declarative-api)
(updated 2026-09-25) documents `agentInvoked` and `respondWith` after
`preventDefault()`. Activation/cancellation events target `document.modelContext`;
window event handlers are unsupported from Chrome 156. Declarative submission and
navigation are not covered by imperative simulation.
As checked 2026-10-12, the CG draft covers the imperative API. [Upstream PR #338](https://github.com/webmachinelearning/webmcp/pull/338),
merged 2026-10-08, removed its declarative section; declarative forms remain an
experimental Chrome implementation with the guide above and a separate explainer.
This documentation update does not extend the dated native evidence below.
## Accepted native evidence: Chrome 150
Governor report: `/tmp/webmcpify-native-browser-governor-report.md`, run
2026-10-05T05:31:43.012Z, Chrome 150.0.7871.186, headed with WebMCP and
WebMCPTesting, temporary synthetic loopback fixture only:
| Probe | Observed result |
|---|---|
| Callback options | Second argument absent; no execution signal received. |
| Caller abort | Caller rejected with AbortError; delayed side effect still ran. |
| Registration abort during execution | Tool disappeared; caller rejected with UnknownError; callback continued and side effect ran. |
| Omitted execute input | TypeError requiring two arguments; callback was not invoked. |
These observations do not establish modern cancellation, omitted-input defaults,
or current-draft result serialization in Chrome 150. Use the existing side-effect-free
input capability probe to select object versus legacy JSON-string input; never
retry a real application tool to discover compatibility.
## Accepted native evidence: Chrome Beta 156
Governor evidence: `/tmp/webmcpify-beta-governor-evidence.json`, run
2026-10-05T05:38:51.840Z, Chrome Beta 156.0.8078.4, headless with WebMCP and
WebMCPTesting, temporary synthetic loopback fixture only. All five probes pass:
| Probe | Observed result |
|---|---|
| Input shape | Object input invoked the callback exactly once. |
| Callback options | Second argument contained an AbortSignal. |
| Caller abort | Started callback received the signal; caller rejected with AbortError; cooperative handler suppressed the delayed side effect. |
| Registration abort during execution | Tool disappeared; already-started callback completed, side effect ran, and caller fulfilled. |
| Omitted execute input | Callback invoked once with `{}`. |
Successful imperative results arrived as serialized JSON strings: the callback
object `{ accepted: true }` produced the string `{"accepted":true}`. `unregisterTool`
was absent; removal was tested only through the registration AbortSignal, not a
distinct unregister operation. These five probes do not establish native WPT,
declarative submission/navigation, or debugging-annotation conformance.
## Local runtime and verification boundaries
Forward `options?.signal` unchanged through wrappers to the fourth argument:
```ts
execute: singleFlight((input, options?: ToolExecuteCallbackOptions) =>
dispatchAndWait('webmcp:save', input, 10000, options?.signal)),
```
The bridge handles preabort without dispatch, cooperative cancellation and cleanup.
Handlers must pass the dispatched signal to asynchronous work and check it before
committing state. Cancellation/timeout leaves an unknown outcome when handlers or
servers continue; inspect state before a manual retry. Registration disposal removes
availability and is not an execution-cancellation mechanism.
`tests/workbench.test.mjs` exercises the read-only Workbench UI in JSDOM with
modeled Native and real Simulated transports, imperative/declarative dispatch,
forced-click/direct-dispatch attempts, exclusions and successful reads. It does
not prove native API conformance. `tests/workbench-cli.test.mjs` exercises separate
package/browser consent using isolated fake runtimes, without engine downloads.
See `tests/README.md` for prerequisites and commands. The TS/JS mutation journal
helpers remain host-side prerequisites for mutations; neither journaling nor
cancellation guarantees exactly-once effects. No continuation-token support is implemented here.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |