dom-snapshot

Browser-free DOM inspection for agents. Fetches or parses server-rendered HTML and returns a compact, structured text snapshot covering headings, ARIA landmarks, navigation, forms, interactive buttons, tables, images, and visible body text — sized to fit comfortably in an LLM context window. Use when: (1) understanding a page's structure before deciding the next action, (2) checking whether a form, heading, or link is present after a navigation event, (3) verifying UI state in a test harness without spinning up Playwright or Selenium, (4) parsing raw HTML retrieved by a previous step. Triggers on: inspect page, DOM snapshot, page structure, check form, check heading, check link, UI state, parse HTML, page metadata, navigation links, accessible landmarks.

Jump to install

Repository: bowen31337/agent-harness-skills
Last source activity: May 4, 2026 at 02:00
Detected SKILL.md language: English
Stars: 3
Forks: 1

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

A direct command skips the review prompt. Inspect the source before running it.

npx skills add https://github.com/bowen31337/agent-harness-skills --skill dom-snapshot

The command stays on one line. Scroll horizontally to inspect it before copying.

Prefer a local copy? Download the files currently available to SkillsMP.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer

2 files

SKILL.md

Source instructions · Read-only preview

Back to search Back to installation

harness-cli

Drive the `harness` Python CLI — the agent-harness-skills toolkit that generates per-project harness configs, runs quality gates, manages execution plans, boots isolated app instances, and emits structured telemetry. Use whenever the user (or another agent) asks to scaffold/refresh harness artifacts, gate a worktree's code against architecture or principles rules, plan or resume a multi-step task, boot an isolated server for an agent worktree, capture screenshots or logs, search the symbol index, or chain any of the above into one invocation with `--then`. Triggers on: harness create, harness lint, harness evaluate, harness plan, harness resume, harness status, harness boot, harness observe, harness screenshot, harness search, harness coordinate, harness audit, harness manifest, harness telemetry, harness completion-report, harness context, harness update, run a harness command, agent harness, agent-harness-skills, quality gate, exec plan, worktree boot, --then pipeline.

2026-05-13

context-handoff

Context handoff protocol for multi-session agent work. When an agent session ends, write a structured summary to the plan progress log so the next agent can continue seamlessly. The handoff contains search hints — specific file paths, grep patterns, glob patterns, and symbol names — that let the next agent rebuild its own context using Read/Grep/Glob tools, rather than relying on pre-assembled content dumps. Use when: (1) ending an agent session where work continues, (2) starting work previously begun by another agent, (3) writing or reading session handoffs, (4) resuming from a previous agent session, (5) implementing multi-agent context continuity, (6) approaching context window limits mid-task. Triggers on: write handoff, session ending, context handoff, hand off context, resume from previous session, pick up where we left off, context window full.

2026-05-04

error-aggregation

Error aggregation view for agents. Groups recent harness error events by domain and frequency so agents can query the error landscape without parsing raw logs. Pre-computes deduplication, trend detection (rising/falling/stable), and JSON-serialisable summaries. Use when: (1) diagnosing which component is failing most often, (2) investigating a spike in a specific domain, (3) checking whether an error rate is rising or recovering, (4) providing a structured error briefing before a remediation agent runs, (5) summarising errors across a multi-agent pipeline. Triggers on: show me recent errors, what errors are happening, error summary, query errors by domain, which domain is failing, error aggregation, rising errors, error frequency.

2026-05-04

View all from this repository

name

dom-snapshot

description

DOM Snapshot Skill

Overview

The DOM Snapshot skill gives agents a browser-free window into page structure. It uses requests + BeautifulSoup to parse server-rendered HTML and emits a compact, labelled text block that fits comfortably inside an LLM context window.

Sections returned:

Section	What it contains
Page Metadata	URL, `<title>`, meta description, `lang` attribute
ARIA Landmarks	Explicit `role=` attributes + implicit landmarks (`<main>`, `<nav>`, …)
Headings	All `<h1>`–`<h6>` in document order
Navigation Links	Links inside `<nav>` elements (or all `<a>` when no `<nav>` is found)
Forms	Each `<form>` with its fields (type, name, label, required) and buttons
Interactive Buttons	Standalone `<button>` elements and `role=button` nodes outside forms
Data Tables	Caption, headers, row count, and up to 3 sample rows per `<table>`
Images	`src`, `alt`, and optional `width`×`height` for up to 10 `<img>` tags
Visible Text	First 1 500 characters of stripped body text (scripts/styles removed)
Errors / Warnings	Present only when something went wrong (HTTP error, missing dependency)

Limitation — only server-rendered HTML is visible. Content injected by JavaScript after the initial response will not appear. For SPAs, capture the page source at the right moment and pass it to dom_snapshot_html.

Workflow

Inspect a live URL? → URL mode

Already have the HTML string in memory? → HTML mode

Need a quick CLI one-liner? → CLI usage

Embedding the snapshot inside an agent? → Programmatic usage

CLI Usage

# Snapshot a live URL:
python skills/dom-snapshot/scripts/snapshot_dom.py https://example.com

# Pass raw HTML via stdin:
curl -s https://example.com | \
    python skills/dom-snapshot/scripts/snapshot_dom.py --html -

# Read HTML from a file:
python skills/dom-snapshot/scripts/snapshot_dom.py --html page.html

# Tune timeouts and link limits:
python skills/dom-snapshot/scripts/snapshot_dom.py \
    https://example.com --timeout 30 --max-links 25

URL Mode

Run the following Python snippet inside an agent or harness step:

import sys
sys.path.insert(0, ".")
from harness_skills.dom_snapshot_skill import dom_snapshot_url

result = dom_snapshot_url(
    "https://example.com",
    timeout=15,    # HTTP timeout in seconds
    max_links=15,  # max navigation links shown
)
print(result)

HTML Mode

Use when you already hold the raw HTML (e.g. captured from a Playwright page or retrieved by a previous requests.get call):

import sys
sys.path.insert(0, ".")
from harness_skills.dom_snapshot_skill import dom_snapshot_html

html = "<html><body><h1>Hello</h1><p>World</p></body></html>"

result = dom_snapshot_html(
    html,
    base_url="https://example.com",  # resolves relative hrefs/srcs
    max_links=15,
)
print(result)

Programmatic Usage

For more control, use the lower-level dom_snapshot_utility API directly:

import sys
sys.path.insert(0, ".")
from dom_snapshot_utility import (
    snapshot_from_url,
    snapshot_from_html,
    snapshot_to_text,
    DOMSnapshot,
)

# Fetch and parse
snap: DOMSnapshot = snapshot_from_url("https://example.com", timeout=20)

# Inspect fields directly
print(snap.meta.title)
print([h.text for h in snap.headings])
print([f.id for f in snap.forms])

# Or render as text
print(snapshot_to_text(snap, max_links=15))

Injecting custom session headers (cookies, auth)

import requests
import sys
sys.path.insert(0, ".")
from dom_snapshot_utility.snapshot import snapshot_from_html

session = requests.Session()
session.headers["Cookie"] = "session=abc123"
resp = session.get("https://example.com/dashboard", timeout=15)
snap = snapshot_from_html(resp.text, base_url="https://example.com/dashboard")

from dom_snapshot_utility import snapshot_to_text
print(snapshot_to_text(snap))

Output Format

### Page Metadata
URL      : https://example.com
Title    : Example Domain
Desc     : This domain is for use in illustrative examples...
Lang     : en

### ARIA Landmarks
  [main] role=main
  [navigation] role=navigation

### Headings
H1: Example Domain

### Navigation Links (3 total)
  'Home' -> https://example.com/
  'Docs' -> https://example.com/docs/
  'Blog' -> https://example.com/blog/

### Forms (1 total)
  FORM id='search' action='/search' method='GET'
    [search] name='q' * label='Search' placeholder='Search docs…'
    Buttons: ['Search']

### Interactive Buttons (2 total)
  [button] 'Open menu'
  [button] 'Close dialog' aria='Close'

### Data Tables (1 total)
  TABLE 'Results' — 42 rows
    Headers: ['Name', 'Status', 'Updated']
    Row: ['alpha', 'passing', '2025-01-10']

### Images (1 total)
  <img> alt='Site logo' (200×50) src=https://example.com/logo.png

### Visible Text (first 1 500 chars)
Example Domain
This domain is for use in illustrative examples in documents…

### Errors / Warnings        ← only present when something went wrong
  ⚠  HTTP 404: Not Found

Data Structures

`DOMSnapshot`

The structured object returned by snapshot_from_html / snapshot_from_url.

Field	Type	Description
`meta`	`PageMeta`	URL, title, description, lang.
`landmarks`	`list[AriaRegion]`	ARIA landmark regions.
`headings`	`list[Heading]`	Headings in document order.
`nav_links`	`list[Link]`	Navigation links (truncated to `max_links`).
`nav_links_total`	`int`	Total links found before truncation.
`forms`	`list[Form]`	Forms with fields and buttons.
`buttons`	`list[Button]`	Standalone interactive buttons.
`tables`	`list[TableSnapshot]`	Table summaries with sample rows.
`images`	`list[ImageSnapshot]`	Image metadata (up to 10).
`visible_text`	`str`	Stripped body text (first 1 500 chars).
`errors`	`list[str]`	Any fetch or parse errors encountered.

`Form` / `InputField`

Field	Type	Description
`id`	`str`	Form `id` attribute.
`action`	`str`	Resolved `action` URL.
`method`	`str`	`GET` or `POST`.
`fields`	`list[InputField]`	Visible input fields (hidden/submit excluded).
`buttons`	`list[str]`	Button labels inside the form.

InputField: type, name, label, placeholder, required.

Key Files

Path	Purpose
`dom_snapshot_utility/snapshot.py`	Core implementation — parsing, dataclasses, `snapshot_to_text`.
`dom_snapshot_utility/__init__.py`	Public API exports for the utility package.
`harness_skills/dom_snapshot_skill.py`	Skill wrapper — `dom_snapshot_url`, `dom_snapshot_html`, CLI entry point.
`skills/dom-snapshot/scripts/snapshot_dom.py`	Standalone CLI helper (no install required).
`tests/test_dom_snapshot.py`	Full unit test suite (offline — HTTP mocked).

dom-snapshot

Source facts

Install options

Review the source files

More from this repository

DOM Snapshot Skill

Overview

Workflow

CLI Usage

URL Mode

HTML Mode

Programmatic Usage

Injecting custom session headers (cookies, auth)

Output Format

Data Structures

DOMSnapshot

Form / InputField

Key Files

More from this repository

DOM Snapshot Skill

Overview

Workflow

CLI Usage

URL Mode

HTML Mode

Programmatic Usage

Injecting custom session headers (cookies, auth)

Output Format

Data Structures

DOMSnapshot

Form / InputField

Key Files

`DOMSnapshot`

`Form` / `InputField`

`DOMSnapshot`

`Form` / `InputField`