doc-summariser

Doc Summariser

An agent skill that demonstrates the SKILL.md + helper-code pattern with a real, runnable capability.

Capability

Given a block of text, produce a faithful summary. Two modes are available:

• Extractive summary (summarise_document): selects the most salient whole sentences from the text using a frequency-based scoring scheme (a variant of the classic Luhn method) and returns them in their original order, so the summary reads naturally.
• One-line gist (gist): a word-bounded single line for when the caller wants the shortest possible version.

Both modes are deterministic and run fully offline, which makes the skill cheap, reproducible, and easy to test. They share their implementation with the MCP summarise tool, so there is one source of truth to upgrade later (for example, to a Claude-generated abstractive summary).

When to use

• The user pastes a long passage and asks for the "gist", "TL;DR", or a "short version".
• An orchestrator needs a quick, cheap summary step before deeper processing.

How to use

from agent_skills.doc_summariser.helper import summarise_document, gist

# Most salient sentences, original order:
summarise_document(long_text, max_sentences=3)

# A single short line:
gist(long_text, max_words=20)

Or run the skill directly to see both modes on a sample:

python agent_skills/doc_summariser/helper.py

Inputs

• text (str): the document to summarise.
• max_sentences (int, default 3): upper bound on extractive summary length.
• max_words (int, default 20): upper bound on the one-line gist.

Output

• summarise_document: a string containing the top sentences, joined and in original document order. If the text is already within the sentence bound it is returned (whitespace-normalised) unchanged.
• gist: a single short line; the original text if already short enough, otherwise the first max_words words followed by an ellipsis.

Edge cases

• A negative bound raises ValueError.
• A bound of zero returns an empty string.
• Text with no scorable content falls back to the leading sentences.