How do I install the Gathering Context skill?

Run npx skillmds@latest add arozumenko/gathering-context in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Gathering Context safe to use?

This skill has not completed SkillMD's automated safety review yet. Capability flags: docs only. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Gathering Context?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Gathering Context free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under Apache-2.

Who published Gathering Context?

arozumenko (@arozumenko) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Gathering Context by arozumenko — Agent Skill

Name: Gathering Context
Rating: 4.666666666666667 (3 reviews)
Author: arozumenko

Gathering Context

Use when the user says 'what do we know about X', 'find prior discussions with Y', or before drafting any reply where prior history matters. Gathers cross-channel context (local KB, email, Teams, optional web) about a person or topic before responding.

arozumenko Updated Sep 22, 2026

File contents

Context Gatherer

Find prior discussions, decisions, and commitments across channels — fast. Walk channels in cost order and stop early when you have enough.

Inputs

Person — name or email
Topic — short phrase
Time window — default: last 2 weeks. Widen only if the topic clearly demands history.

Search Order (fast → slow)

1. Local knowledge base

If a local index exists (e.g. .knowledge/threads.json), Read it and Grep for the person's email and topic keywords.
This gives pre-summarized threads — usually enough on its own for routine asks.

2. Email

search-query with from:<email> AND <keywords> (and to:<email> variant).
Fall back to list-mail-messages + manual filter if search-query is unavailable.
Pull bodies with get-mail-message only for the top 3–5 hits — never bulk-read.

3. Teams chats

find-chat by name → list-chat-messages for the matched chats with recent activity.
Aggregate consecutive messages from the same sender within ~2 minutes into one turn before summarizing — people fragment messages and raw dumps are noisy.

4. Web (only if needed)

tavily_search for public topics (announcements, vendor docs). Never web-search internal/private topics.

Stop as soon as you have enough to answer the question. Don't exhaust every channel by default.

Output

Max ~500 words. Facts, not essays.

## Context for <person> — <topic>

### Prior conversations
- [date] <channel>: <one-line summary>
- [date] <channel>: <one-line summary>

### Decisions & commitments
- <what was agreed, by whom, when>

### Related discussions
- <relevant context from other people/threads>

### Open items
- <anything unresolved or pending>

If nothing relevant exists, say so in one line. Never fabricate context.

Rules

Default window: last 2 weeks. Widen only with explicit need.
Summarize — never paste raw email bodies or message dumps.
Mask PII / credentials encountered in searches.
If channels disagree, flag the conflict rather than picking a side.
Aggregate Teams message bursts before summarizing.

---name: gathering-contextdescription: Use when the user says 'what do we know about X', 'find prior discussions with Y', or before drafting any reply where prior history matters. Gathers cross-channel context (local KB, email, Teams, optional web) about a person or topic before responding.license: Apache-2.0---# Context GathererFind prior discussions, decisions, and commitments across channels — fast. Walk channels in cost order and stop early when you have enough.## Inputs- **Person** — name or email- **Topic** — short phrase- **Time window** — default: last 2 weeks. Widen only if the topic clearly demands history.## Search Order (fast → slow)### 1. Local knowledge base- If a local index exists (e.g. `.knowledge/threads.json`), `Read` it and `Grep` for the person's email and topic keywords.- This gives pre-summarized threads — usually enough on its own for routine asks.### 2. Email- `search-query` with `from:<email> AND <keywords>` (and `to:<email>` variant).- Fall back to `list-mail-messages` + manual filter if `search-query` is unavailable.- Pull bodies with `get-mail-message` only for the top 3–5 hits — never bulk-read.### 3. Teams chats- `find-chat` by name → `list-chat-messages` for the matched chats with recent activity.- **Aggregate consecutive messages** from the same sender within ~2 minutes into one turn before summarizing — people fragment messages and raw dumps are noisy.### 4. Web (only if needed)- `tavily_search` for **public** topics (announcements, vendor docs). Never web-search internal/private topics.Stop as soon as you have enough to answer the question. Don't exhaust every channel by default.## OutputMax ~500 words. Facts, not essays.```## Context for <person> — <topic>### Prior conversations- [date] <channel>: <one-line summary>- [date] <channel>: <one-line summary>### Decisions & commitments- <what was agreed, by whom, when>### Related discussions- <relevant context from other people/threads>### Open items- <anything unresolved or pending>```If nothing relevant exists, say so in one line. **Never fabricate context.**## Rules- Default window: last 2 weeks. Widen only with explicit need.- Summarize — never paste raw email bodies or message dumps.- Mask PII / credentials encountered in searches.- If channels disagree, flag the conflict rather than picking a side.- Aggregate Teams message bursts before summarizing.

Frequently asked questions