replay-incident-risk

name	replay-incident-risk
description	Cross-reference a posthog-js or rrweb change against past incident patterns before writing, reviewing, or merging it. Use when a diff touches session recording, the recorder or any lazy-loaded extension bundle (entrypoints/), session rotation or idle handling, fetch/XHR wrappers, rrweb record or replay code, recording triggers or remote config, or release workflows. Also use when asked "is this change risky", "could this repeat an incident", or "will this affect recording volume".
compatibility	Requires a posthog-js checkout with git history (origin/main reachable).

Replay incident risk review

Changes to posthog-js and the rrweb fork have repeatedly caused severe incidents: fleet-wide recording loss, over-recording that inflated bills, broken customer sites, silent replay corruption, and a stored XSS. Most repeated through a small set of code-level patterns. This skill checks a diff against those patterns.

The registry of patterns and incidents lives next to this file: INCIDENTS.md. Read the matching class sections, not the whole file.

Workflow

1. Run the mechanical matcher from the repo root:

   node .agents/skills/replay-incident-risk/check.mjs [base-ref] [head-ref]
   

   It diffs HEAD against origin/main by default and prints matched incident classes with the touched paths and risky added lines. It is advisory and always exits 0, even on internal errors.

   When changing risk-map.json itself, also run check.mjs --validate: it fails on path patterns matching no tracked file and on anchors that don't resolve to an INCIDENTS.md heading, so dead patterns can't rot silently. CI runs it on every PR.