libretto 0.5.0 → 0.5.1

package/skills/libretto/references/reverse-engineering-network-requests.md

@@ -1,39 +0,0 @@
-# Reverse Engineering Network Requests
-
-Use this reference when the user wants to turn a browser workflow into direct network requests.
-
-This is the default approach for new integrations when the site exposes a clear and stable HTTP request path.
-
-## When to Use This
-
-- The page clearly loads or submits data through HTTP requests.
-- The user can perform the workflow manually in a headed browser.
-- Replaying the request is likely faster or more stable than reproducing every UI action.
-- Fall back to browser automation when the request path is unclear, too dynamic, or blocked by anti-bot systems.
-
-## Workflow
-
-- Open the page in headed mode.
-- Let the user perform the relevant workflow manually.
-- Read the network log after the relevant step.
-- Identify the smallest set of requests that actually carries the data or performs the action.
-- Confirm with the user before replaying any request that could mutate data.
-- Recreate the request in code outside Libretto.
-- Verify the resulting workflow with `npx libretto run ...`.
-
-## Commands
-
-```bash
-npx libretto open https://target.example.com --headed
-npx libretto network --last 20
-npx libretto network --method POST --last 20
-npx libretto network --filter 'referral|patient|search'
-npx libretto exec "return await networkLog({ method: 'POST', last: 10 })"
-```
-
-## Notes
-
-- Start with the request that returns the data you need, not every request on the page.
-- Prefer captured requests over guessing payload shape.
-- If the request format is opaque, highly dynamic, or heavily defended, fall back to UI automation for that part.
-- Treat all replayed requests as potentially side-effectful until proven otherwise.

package/skills/libretto/references/user-action-log.md

@@ -1,31 +0,0 @@
-# User Action Log
-
-Use this reference when the user performs steps manually in a headed Libretto session and you want to incorporate those steps into a workflow.
-
-## When to Use This
-
-- The user demonstrates the workflow in the browser window.
-- You want to know what they clicked, typed, or selected.
-- You need to reconcile manual user actions with captured network requests.
-
-## Workflow
-
-- Open the session in headed mode.
-- Ask the user to perform the workflow.
-- Read the user action log after they finish.
-- Use the log to identify the important transitions in the workflow.
-- Combine the action log with `snapshot` or `network` when the log alone is not enough.
-
-## Commands
-
-```bash
-npx libretto actions --source user --last 20
-npx libretto actions --source user --filter 'button|input|select'
-npx libretto exec "return await actionLog({ source: 'user', last: 10 })"
-```
-
-## Notes
-
-- The action log is most useful for reconstructing the sequence of a workflow, not for discovering every selector you need.
-- If the user performed relevant manual steps, read the action log before writing or revising the workflow code.
-- Use the action log to anchor your understanding, then inspect the current page state with `snapshot` or `exec`.