Shortread: long content cut to what matters

Manual summarization wastes time and often misses the core signal in a sea of filler. This skill compresses long-form content, transcripts, and codebases into structured, high-density briefs without editorial fluff.

What it does

• Multi-source compression converts documents, URLs, transcripts, or entire directory structures into structured summaries.
• Priority ranking organizes key points by objective importance rather than their order of appearance in the source.
• Actionable extraction lists concrete tasks with owners and drops vague ones like "discuss further".
• Decision logging captures specific outcomes and open questions to prevent context loss in long threads.
• Verbatim preservation ensures that dates, percentages, and currency values remain exact to prevent data drift.

How it works

1. Target selection by passing a file path, URL, or pasted text to the command.
2. Mode adjustment using flags like --brief, --full, or --outline to control the depth of information.
3. Structured parsing where the agent identifies direct quotes, decisions, and action items according to the source content.

Frameworks & tools

This tool is language-agnostic and works with Markdown, plain text, and source code. It integrates with existing project contexts by reading README files and manifests to maintain consistent terminology.

Why this beats prompting it yourself

Generic prompts often lead to polite filler and "this document discusses" introductions that waste vertical space. This skill enforces strict token discipline and uses specific data extraction rules to ensure every line provides information, not just a description of information.

Use cases

• Summarize 60-minute meeting transcripts into owner-assigned action items.
• Generate high-level overviews of new codebases for rapid onboarding.
• Distill long technical PRDs into brief TLDRs for executive stakeholders.
• Extract core decisions and unresolved questions from long Slack or Discord threads.

Known limitations

The skill requires the target content to be accessible within the agent's current environment or via provided URL. Length limits are subject to the underlying model's context window.