ld-migration-metrics - SKILL.md Agent Skill

name: ld-migration-metrics description: Retrieve migration flag metrics (consistency, latency, error rate) from LaunchDarkly by navigating the dashboard with Playwright. Use when the user wants to check migration flag health, consistency rates, or compare old vs new system performance for any LD migration flag. allowed-tools: Bash(playwright-cli:*), Read, Grep

LaunchDarkly Migration Metrics Skill

Extracts migration flag metrics from the LaunchDarkly dashboard using Playwright browser automation. The LD public API does not expose consistency metrics for migration flags, so this skill scrapes them from the UI.

Prerequisites

Playwright MCP Bridge extension must be installed in the user's browser (Chrome/Edge)
- Install from: https://chromewebstore.google.com/detail/playwright-mcp-bridge/mmlmfjhmonkocbjadbfplnigmagldckm
The user must be logged in to LaunchDarkly in their browser
The user's browser must be open

Input

The skill accepts these arguments (space-separated):

Flag key (required) - e.g., my-flag
Project key (required) - e.g., my-project
Environment (optional, default: production) - e.g., local, qa, staging

Example invocations

/ld-migration-metrics my-flag
/ld-migration-metrics my-flag my-project local
/ld-migration-metrics my-flag my-project staging

Workflow

Follow these steps exactly:

Step 1: Parse arguments

Parse the arguments from the skill invocation:

Arg 1: FLAG_KEY (required - if missing, ask the user)
Arg 2: PROJECT_KEY (required - if missing, ask the user)
Arg 3: ENVIRONMENT (default: production)

Construct the URL:

https://app.launchdarkly.com/projects/{PROJECT_KEY}/flags/{FLAG_KEY}/targeting?env={ENVIRONMENT}&selected-env={ENVIRONMENT}

Step 2: Connect to browser via extension

playwright-cli open --extension

If this fails with "Extension connection timeout", inform the user they need to:

Install the Playwright MCP Bridge extension
Have their browser open

Step 3: Navigate to the flag page

playwright-cli goto {URL}

After navigation, check the page title:

If the title is "Sign in", inform the user they need to log in to LaunchDarkly in their browser first, then close the browser and stop.
If the title contains the flag name, proceed.

Step 4: Take a snapshot and save it

playwright-cli snapshot

Read the snapshot file to get the full page structure.

Step 5: Extract metrics from the snapshot

The snapshot YAML contains the flag's targeting rules and migration metrics. The structure follows this pattern for each rule:

- generic "{RuleName}" [ref=...]: # Rule container with name
    - heading "{RuleName}" [level=3] # Rule heading
    - generic "If cache key is one of..." # Rule clause
    - generic [ref=...]: # Metrics container
        - generic [ref=...]:
            - generic [ref=...]: { value } # Metric value (e.g., "99%", "-")
            - generic [ref=...]: Average consistency # Metric label
        - generic [ref=...]:
            - generic [ref=...]: { value } # Metric value (e.g., "- slower", "14% faster", "-")
            - generic [ref=...]: p99 latency # or "Difference between p99s"
        - generic [ref=...]:
            - generic [ref=...]: { value } # Metric value (e.g., "- higher", "-")
            - generic [ref=...]: Average error rate # Metric label
        - generic [ref=...]:
            - generic [ref=...]: Last 7 days # Time period

To extract the data:

Search the snapshot for all lines containing Average consistency, p99 latency, Difference between p99s, and Average error rate
For each match, the metric value is on the line immediately before the label line
Match each group of 3 metrics to their parent rule by finding the nearest preceding heading with [level=3]

There is also a top-level "Consistency rate" section before the rules. Look for:

- heading "Consistency rate" [level=3]
  ...
  - text: {value}
  - paragraph: Average consistency

Metric values to expect:

Average consistency: A percentage like 99%, 87%, 65%, or "-" if no data
p99 latency: Values like "- slower", "- higher", "14% faster", "7276%slower", "17ms", or "-" if no data
Difference between p99s: Same as p99 latency, just a different label format
Average error rate: Values like "- higher", "- lower", or "-" if no data

Rules that are set to off variation (not in shadow/live mode) will typically show "-" for all three metrics since no dual-read comparison is happening.

Step 6: Identify rule variations

For each rule, also extract the current rollout. Look for buttons within the rule section:

- button "shadow 100%" [ref=...] # variation name and percentage
- button "off 0%" [ref=...]

The active variation is the one with a non-zero percentage.

Step 7: Close the browser

playwright-cli close

Step 8: Present the results

Output a markdown table with the following format:

## Migration Metrics: {FLAG_KEY} ({ENVIRONMENT})

_Last 7 days_

| Rule           | Variation   | Avg Consistency | p99 Latency | Avg Error Rate |
| -------------- | ----------- | :-------------: | :---------: | :------------: |
| **{RuleName}** | {variation} |   **{value}**   |   {value}   |    {value}     |
| ...            | ...         |       ...       |     ...     |      ...       |

Formatting rules:

Bold the rule name in the first column
Bold the consistency percentage value
Use - for metrics with no data
Include the top-level overall metrics as the first row labelled "Overall"
Include the default fallthrough as the last row labelled "Default fallthrough"
Sort rules in the order they appear on the page

After the table, add a brief summary highlighting:

Any rules with consistency below 95% (these need attention)
Any rules showing notably higher error rates or slower latency
Rules still on off that have no metrics

Error Handling

If the browser extension connection fails, tell the user to install the extension and retry
If the page shows "Sign in", tell the user to log in to LaunchDarkly first
If the flag page shows "Not found", the flag key or project key may be wrong
If no migration metrics are found in the snapshot, the flag may not be a migration flag