X Actionbook Recap

Collect, scroll, extract, and summarize recent X (Twitter) posts for any handle (optionally filtered by keyword search) using the Actionbook Rust CLI (actionbook-rs) workflow (open → snapshot/accessibility tree → extract `article` text). Use when asked to analyze a handle over a time window (e.g., last 7 days), produce Chinese working notes, and publish a neutral English recap (single post or thread) from a specified account.

Security Scan

Instructions and requirements are internally consistent: the skill is an instruction-only workflow for using actionbook-rs to collect, summarize, and optionally publish X posts; nothing in the bundle asks for unrelated credentials or hidden endpoints.

Purpose & Capability

The skill's described purpose (collect → extract → summarize → publish on X using actionbook-rs) matches the SKILL.md instructions. Minor inconsistency: the runtime assumes the actionbook (actionbook-rs) CLI and browser automation are available, but the registry metadata lists no required binaries or install steps. The absence of a declared required-binary for actionbook-rs should be fixed but does not indicate malicious intent.

Instruction Scope

SKILL.md stays on task: it instructs opening X pages, snapshotting the accessibility tree, extracting article nodes, summarizing (Chinese internal notes + English publish draft), optionally creating an image, and publishing. It does not instruct reading unrelated local files or environment variables. It does include publishing steps and explicitly requires user confirmation before posting, which mitigates the risk of accidental posting.

Install Mechanism

This is instruction-only with no install spec or code files, so there is no automated download or execution of third-party code. The only risk is that it presumes external tooling (actionbook-rs and a browser automation environment) which must be installed separately by the operator.

Credentials

The skill declares no environment variables or credentials, which is proportionate. Practical publishing will require a logged-in browser session or X account credentials under the agent's control; those are not requested by the skill but are necessary for the publish step. The skill does not attempt to exfiltrate secrets or require unrelated credentials.

Persistence & Privilege

No elevated privileges or persistent install behavior are requested. always:false and default model-invocation settings are used. The skill does not attempt to modify other skills or system settings.

Assessment

This skill is an instruction-only workflow for using actionbook-rs to scrape and summarize public X posts and optionally post a recap. Before installing/using: (1) ensure you have a trusted installation of actionbook-rs and any browser automation tooling the skill will rely on (the skill assumes these but doesn't declare them); (2) know that publishing requires a logged-in browser or account — verify and confirm every post before posting (the SKILL.md asks for confirmation, but you should enforce it); (3) consider X/TOS and privacy/legal implications of scraping and reposting content; (4) because the skill is from an unknown source, only run its instructions in a controlled environment and avoid giving it access to sensitive or unrelated browser sessions; (5) if you want the agent to be unable to post autonomously, restrict model-invocation or require explicit human approval when the agent attempts any publish action.

X recap via actionbook-rs

What this skill is for

Produce a repeatable “collect → extract → summarize → publish” workflow for any X handle (optionally with a keyword) using the actionbook-rs approach:

1. actionbook browser open the profile/search page
2. actionbook browser snapshot to get the accessibility tree (incl. article nodes)
3. (optional) actionbook browser eval to scroll
4. extract post text from article blocks
5. analyze + draft output (Chinese internal notes; English publish)
6. publish on X (neutral tone; optionally attach an image)

Guardrails

• Infinite scroll is not exhaustive; be explicit about coverage limits.
• Don’t quote “recent interviews” unless the user provides exact links/timestamps.
• Publishing is external action: confirm the target account + final copy before posting.

Workflow

1) Collect posts (Actionbook)

Pick one entry point:

• Profile: https://x.com/<handle>
• Search (keyword + optional recency): https://x.com/search?q=from%3A<handle>%20<keyword>&src=typed_query&f=live

Commands (example):

# open (profile)
actionbook browser open "https://x.com/<handle>"

# snapshot (repeat after each scroll)
actionbook browser snapshot --refs aria --depth 18 --max-chars 12000

# scroll a bit
actionbook browser eval "window.scrollBy(0, 2200)"

Extraction heuristic:

• In snapshots, locate article nodes that contain the post text.
• Record for each post:
  • text (verbatim)
  • timestamp shown (relative or absolute)
  • URL if present
  • whether it’s a repost/quote (note it)

Stop condition:

• You have enough coverage for the user’s time window (e.g., 7 days) OR diminishing returns.

2) Summarize (Chinese notes)

Write a compact Chinese working summary:

• themes (3–6 bullets)
• representative posts (links)
• what’s missing / uncertainty

3) Draft publish copy (English, neutral)

Choose output type:

• Single post (≤280 chars) OR
• Thread (6–10 parts) if needed

Use neutral framing:

• “Observation from public posts …”
• avoid mind-reading; separate “what he said” from interpretation

Use templates in references/templates.md.

4) Image (optional but recommended)

Preferred options (no Python required):

• Clean crop of the relevant post (browser screenshot at 1280×720 + zoom)
• A simple HTML/SVG card rendered in browser and screenshotted (see references/image-card.md)

5) Publish on X

If using OpenClaw browser automation:

• open compose
• paste final English copy
• upload image (if any)
• post / thread

Confirm before posting:

• target handle (e.g., @gblwll)
• final text
• image choice

Bundled references

• references/templates.md — recap + thread templates (English)
• references/checklist.md — extraction checklist + caveats
• references/image-card.md — HTML/SVG card approach (no Pillow)