skillZs
★ LIVE SKILL TAGS ★
>>> LIVE SKILLS INDEX <<<
* OPEN SOURCE *
NO LOGIN, NO TRACKING
※ REAL INSTALL DATA ※
← back to all skills
us/crw60 installs

crw-watch

Detect what changed between two page snapshots with fastCRW — stateless diff as a REST primitive. Use when you need to track content changes, monitor a page for updates, or build a cron-based alert system: "has this page changed?", "alert me when pricing changes", "diff this week's scrape against last week's". Step 7 of the crw workflow ladder.

How do I install this agent skill?

npx skills add https://github.com/us/crw --skill crw-watch
view source ↗

Is this agent skill safe to install?

  • Gen Agent Trust Hubpass

    This skill provides a change-tracking and diffing system using the fastCRW API. It ingests untrusted web content which presents an indirect prompt injection attack surface, although the provided examples demonstrate secure practices for handling and escaping this data using standard shell utilities.

  • Socketpass

    No alerts

  • Snykwarn

    Risk: MEDIUM · 1 issue

What does this agent skill do?

crw-watch — change tracking and diffing

When to use

  • You want to know what changed between two snapshots of a page.
  • Step 7 in the crw ladder. Assumes you can already scrape the page — see crw-scrape (step 2).
  • You want a self-hosted, stateless diff primitive you control. Firecrawl offers change tracking only as a managed cloud feature; crw exposes the same primitive as a REST endpoint that runs on your own infra — you own the snapshots, the cadence, and the data.

Architecture: crw is stateless

crw stores nothing between calls. The caller owns the snapshots:

1. Scrape now        → store snapshot (markdown / json)
2. Scrape later      → call /v1/change-tracking/diff with current + previous
3. On status=changed → alert / act
4. Repeat on a cron

Diff modes

Two modes, composable:

ModeWire stringWhat it produces
Git-style text diff"gitDiff" (alias: "git-diff")Unified-diff text + parse-diff AST in diff.text / diff.json
Per-field JSON diff"json"Path-keyed map {"$.field": {"previous":…,"current":…}} in diff.json; requires schema

Default (omit modes): ["gitDiff"]. Combine both: "modes": ["gitDiff", "json"].

Quick start

Single page diff (REST):

curl -X POST "$CRW_API_URL/v1/change-tracking/diff" \
  -H "Authorization: Bearer $CRW_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "current": {
      "markdown": "# Pricing\nPro plan: $49/mo"
    },
    "previous": {
      "markdown": "# Pricing\nPro plan: $39/mo",
      "contentHash": "<hash from prior result>"
    },
    "modes": ["gitDiff"]
  }'

Response shape:

{
  "success": true,
  "data": {
    "status": "changed",
    "firstObservation": false,
    "contentHash": "<new hash>",
    "snapshot": { "markdown": "...", "contentHash": "..." },
    "diff": {
      "text": "@@ -1,2 +1,2 @@\n # Pricing\n-Pro plan: $39/mo\n+Pro plan: $49/mo",
      "json": { "files": [...] }
    }
  }
}

Batch diff (discriminated by presence of batch key):

curl -X POST "$CRW_API_URL/v1/change-tracking/diff" \
  -H "Authorization: Bearer $CRW_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "batch": [
      { "url": "https://example.com/pricing", "current": {"markdown": "..."}, "previous": {"markdown": "..."} },
      { "url": "https://example.com/about",   "current": {"markdown": "..."} }
    ],
    "modes": ["gitDiff"]
  }'

Shared modes/schema/prompt/contentType at the top level are defaults; each batch item can override them individually.

Inline during a scrape — pass changeTracking as a format on /v1/scrape:

curl -X POST "$CRW_API_URL/v1/scrape" \
  -H "Authorization: Bearer $CRW_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://example.com/pricing",
    "formats": ["markdown", "changeTracking"],
    "changeTracking": {
      "modes": ["gitDiff"],
      "previous": { "markdown": "...", "contentHash": "..." }
    }
  }'

Request fields

Single mode: { current, previous?, modes, schema?, prompt?, contentType?, tag?, goal?, judgeEnabled? }

Batch mode: { batch: [...items], modes, schema?, ... } where each item is { url?, current, previous?, modes?, schema?, ... }.

FieldTypeNotes
current.markdownstringCurrent page content (gitDiff / mixed)
current.jsonobjectCurrent extracted JSON (json / mixed)
previous.markdownstringPrior snapshot for gitDiff
previous.contentHashstringPersist from prior result's snapshot.contentHash
modesstring[]["gitDiff"] (default), ["json"], or both
schemaJSON SchemaRequired for json mode; defines tracked fields
promptstringNatural-language extraction prompt (alternative to schema)
contentTypestringIf binary/non-text, triggers byte-hash comparison only
tagstringOpaque caller ID echoed back on the result
goalstringNatural-language filter for meaningful changes (AI judge, M2)
judgeEnabledboolEnable AI judgment (M2 feature; accepted but not yet applied)

The goal field (AI judge)

goal is a natural-language filter for what counts as a meaningful change, fed to an LLM judge. It is accepted by the server now but applied in a future milestone (M2). Guidance for when it lands:

  • Be specific: "Alert when the listed price changes; ignore copy rewrites and nav updates" beats "detect important changes".
  • Narrow the scope: "Only flag changes to the Features table, not the hero section".
  • The judge returns {meaningful, confidence, reason, meaningfulChanges[]} in the result's judgment field.

Cron pattern (self-hosted)

#!/usr/bin/env bash
# cron-check.sh — run every hour via cron or a scheduler
SNAPSHOT_FILE=".crw/snapshot.json"
CURRENT=$(crw scrape "https://example.com/pricing" --format markdown)

if [ -f "$SNAPSHOT_FILE" ]; then
  PREV_MARKDOWN=$(jq -r '.markdown' "$SNAPSHOT_FILE")
  PREV_HASH=$(jq -r '.contentHash' "$SNAPSHOT_FILE")
  RESULT=$(curl -s -X POST "$CRW_API_URL/v1/change-tracking/diff" \
    -H "Authorization: Bearer $CRW_API_KEY" \
    -H "Content-Type: application/json" \
    -d "{\"current\":{\"markdown\":$(jq -Rsc . <<<"$CURRENT")},\"previous\":{\"markdown\":$(jq -Rsc . <<<"$PREV_MARKDOWN"),\"contentHash\":\"$PREV_HASH\"},\"modes\":[\"gitDiff\"]}")
  STATUS=$(echo "$RESULT" | jq -r '.data.status')
  if [ "$STATUS" = "changed" ]; then
    echo "CHANGED: $(echo "$RESULT" | jq -r '.data.diff.text')"
    # → send alert, write to DB, trigger webhook, etc.
  fi
  echo "$RESULT" | jq '.data.snapshot' > "$SNAPSHOT_FILE"
else
  # First observation — store the snapshot
  curl -s -X POST "$CRW_API_URL/v1/change-tracking/diff" \
    -H "Authorization: Bearer $CRW_API_KEY" \
    -H "Content-Type: application/json" \
    -d "{\"current\":{\"markdown\":$(jq -Rsc . <<<"$CURRENT")},\"modes\":[\"gitDiff\"]}" \
    | jq '.data.snapshot' > "$SNAPSHOT_FILE"
fi

Tips

  • Persist snapshot from each result as the next call's previous. The snapshot field in the response contains the normalized content and contentHash — store it, don't recompute it.
  • firstObservation: true means no previous was supplied. The server sets status: "changed" and returns snapshot but produces no diff. Store it as your baseline.
  • json mode needs current.json (+ optionally a schema). Without structured input it produces no diff — use gitDiff mode for plain markdown.
  • Batch is more efficient at scale. One HTTP round-trip for N pages instead of N calls. Top-level modes/schema as defaults keeps the body compact.
  • Data sovereignty. You supply previous; crw computes and returns. Nothing is stored server-side. Your snapshots, your infra, your retention policy.

See also

  • crw-scrape — get the current page content to feed into the diff
  • crw — ladder overview and routing rules

Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.

<a href="https://skillzs.dev/skills/us/crw/crw-watch">View crw-watch on skillZs</a>