3080 Brief

30-second judgment, 80% in one picture

Chinese (Simplified) · Quick start · Worked example · Latest release

An open-source AI Agent Skill that turns source documents into reader-first decision briefs: a 30-second executive summary, one auditable visual, a key-question table, and a clear storyline. It preserves the source and can create a new brief in Feishu/Lark, Markdown, or Word/docx when the host supports that format.

Real output example: Anthropic's five-minute article explains token pricing, prompt caching, context growth, session length, and subagents across multiple sections. The 3080 Brief turns it into one operating judgment, one five-step routine, and a clear order for what to check first, while keeping the original as the source of truth. Open the complete example →

What you get

The source remains unchanged; the skill creates a separate brief.

Quick start

1. Install the complete Skill folder

Ask any Agent Skills-compatible agent:

Install the Agent Skill from https://github.com/BobbyYue/3080-brief,
using the complete subdirectory skills/3080-brief. Register it in your normal Skill directory,
then check the dependencies required for my target output and ask once before installing anything missing.

Or install it manually:

git clone --depth 1 https://github.com/BobbyYue/3080-brief.git
cp -R ./3080-brief/skills/3080-brief "<YOUR_AGENT_SKILLS_DIR>/3080-brief"

Copy the entire folder—not only SKILL.md—and reload the agent if its documentation requires it.

2. Give it a source and a concrete request

Use $3080-brief to turn this document into a new reader-first decision brief.
Keep the source unchanged and create the result in the same format.
Return the generated output link or file, plus the final verification status.

Natural language works too:

Create a new brief from this proposal: make the conclusion clear in 30 seconds,
cover the core logic in one picture, and answer the reader's key questions in one table.

Why use it

• Reader-first: reorganizes material around what readers need to understand, trust, decide, or do.
• Core-value extraction: separates decision-critical conclusions, evidence, risks, and actions from background noise.
• Clear storyline: turns fragmented or technical source logic into a coherent argument.
• Visual reasoning: uses one auditable picture to reveal the dominant relationships rather than decorate the page.
• Reliable delivery: preserves the source, traces important claims, matches the output format, and verifies the final artifact.

Installation options

Send this complete instruction to an agent that can install Skills from GitHub:

Install the Agent Skill from https://github.com/BobbyYue/3080-brief,
using the complete subdirectory skills/3080-brief and register it in your Skill registry.
Then run the dependency check for my target output, show one complete plan for every missing dependency,
and ask once whether to install or enable all of them. If I approve, handle every listed item without
separate per-item approvals. For Feishu output, register beautiful-feishu-whiteboard as an independent
Skill and enable executable lark-doc and lark-whiteboard workflows when missing. Loading a Skill
specification does not count as executable readiness. If I decline, keep 3080-brief installed but leave
the affected output path blocked.

git clone --depth 1 https://github.com/BobbyYue/3080-brief.git
Copy-Item -Recurse ./3080-brief/skills/3080-brief "<YOUR_AGENT_SKILLS_DIR>/3080-brief"

Download the latest release, extract it, and import skills/3080-brief through the client's Skill UI. Do not upload the repository root unless the client explicitly supports repository subpaths.

3080-brief follows the open Agent Skills folder format. Client commands, registry paths, reload behavior, script execution, and document integrations vary by host; see the official client showcase.

Compatibility and assurance

The fixed production path is: initialize → freeze source evidence → preflight → create the full draft → review the recorded artifact → re-fetch source and output → finalize. A plan or acknowledgement is never the final deliverable.

Standard and Strict require three independent review executions. Fast may use three role-separated self-checks only when the user explicitly requests Fast.

Core offline validation requires Python 3.9+ and no third-party Python packages. Feishu/Lark output additionally requires:

• executable lark-doc read/create and lark-whiteboard query/update workflows;
• Node.js 20+;
• @larksuite/cli / lark-cli 1.0.60+;
• @larksuite/whiteboard-cli exactly 0.2.11 in the isolated tool cache;
• beautiful-feishu-whiteboard 1.1.1+;
• Feishu/Lark authentication and the required document permissions.

Missing Feishu dependencies block only that output path. The skill shows the source, version, known effects, and proposed action before requesting approval. It does not silently install software or grant account permissions.

Development verification

Run the complete offline suite:

bash skills/3080-brief/scripts/self_test.sh

Useful focused checks:

python3 skills/3080-brief/scripts/check_context_budget.py skills/3080-brief --json
python3 skills/3080-brief/scripts/check_dependencies.py --mode feishu --json

Repository layout:

skills/3080-brief/   installable Agent Skill
docs/examples/       source-linked worked examples
docs/assets/         README and social-preview visuals
.github/workflows/   offline CI

Privacy and limitations

• Never publish source document tokens, tenant identifiers, credentials, or internal metrics in issues or examples.
• Independent review and blind-reader claims are made only when those capabilities actually ran.
• Source-linked examples reflect the product behavior and evidence available on their stated retrieval date.
• This repository does not use with-skill/without-skill benchmark claims.

See CONTRIBUTING.md and SECURITY.md. Licensed under MIT.