colleague-distillation
Skill472 starsChanged 8 months agoMITView on GitHub
Distill a colleague into a reusable AI skill (work + persona) using tool connections — Slack, Slack AI, Jira, GHE, Bitbucket, Confluence, SharePoint, Teams, Outlook, Notion, Linear, Google Docs, and more — without manual paste. Use when the user wants a colleague skill, digital twin of a coworker, or capture of someone's technical voice from workplace systems. Requires tool_connections + 10xProductivity verified_connections (or equivalent .env).
- Reads credentials
In this repository11 other files it gives its agents.
- assistant-orchestratorSkill
- colleague-distillationSkill
- create-workflowSkill
- discover-ui-surfaceSkill
- linkedin-engagementSkill
- assistant-orchestratorSkill
- create-workflowSkill
- discover-ui-surfaceSkill
- enterprise-searchSkill
- linkedin-engagementSkill
- tool-connectorSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 41
Use the same slug and folder layout in both.
- Distinctive: line 55
Before searching, pin who the colleague is:
- Very specific: line 108
Persona (
persona.md): Use a layered structure compatible with colleague-skill: - Distinctive: line 116
Grounding: Prefer quoted paraphrases with source pointers.
- Very specific: line 171
Do not exfiltrate secrets, PII bundles, or restricted content into colleagues/ — redact tokens, customer data, and health/financial …
- Distinctive: line 172
When unsure whether content is allowed in a repo, ask the user before writing.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
10xProductivity skill: This file is the Cursor entry point for colleague distillation when working in this repo.
Colleague distillation (tool-backed)
Purpose
Produce a colleague skill: structured work knowledge (systems, standards, review style) plus persona (tone, decisions, interpersonal habits), using APIs and search already wired in tool_connections / 10xProductivity — not hand-pasted exports.
Output layout matches the open colleague-skill convention so results can coexist with that generator:
colleagues/{slug}/work.mdcolleagues/{slug}/persona.mdcolleagues/{slug}/meta.jsoncolleagues/{slug}/SKILL.md(merged invocable skill)
Optional: Clone colleague-skill for its prompts/work_analyzer.md, persona_analyzer.md, work_builder.md, persona_builder.md if you want identical extraction templates; this skill defines what to fetch and where to write.
Prerequisites
- Load the relevant 10xProductivity connection docs under
tool_connections/and any allowed private recipes under$TENX_PRIVATE_DIR/personal/. - Load
$TENX_PRIVATE_DIR/verified_connections.md— only call tools listed there (or documented in10xProductivity/tool_connections//personal/). - For Jira, use the verified Jira connection documentation and recipes in this repo.
- Credentials:
sourceor load$TENX_PRIVATE_DIR/.env(or project.env) beforecurl/ scripts. Never commit secrets.
Cursor vs Claude Code
| Environment | Where to put generated files | How this skill is loaded |
| Cursor | Repo root: colleagues/{slug}/ | .cursor/skills/colleague-distillation/SKILL.md |
| Claude Code | Same colleagues/{slug}/ under the active project | .claude/skills/colleague-distillation/SKILL.md |
Use the same slug and folder layout in both; only the skill install path differs.
Slug rules
- Slug = unique directory name:
michael_donnelly,michael_donnelly_2, … (ASCII, underscores). - Collisions: Same slug overwrites an existing colleague folder. Disambiguate with
_2,_3, or a distinct codename. - Store display name and aliases in
meta.json, not only in the slug.
Phase 1 — Resolve identity
Before searching, pin who the colleague is:
- Active Directory (if configured in tool_connections): resolve email, manager chain, department — use for Jira/Slack account mapping when IDs are unknown.
- Slack: From
verified_connections.md, use Slack API recipes to resolve@handle→ user id (U…) forfrom:@user/from:U…search syntax. - Jira: Resolve accountId (assignee, reporter, comment author) via Jira user search API — see
jiraskill.
Record: slack_user_id, jira_account_id, email, ad_cn (as available).
Phase 2 — Pull source material (priority order)
Gather raw excerpts (save under colleagues/{slug}/knowledge/raw/ as .md or .json snippets) with source + URL/ticket/channel + date in each chunk header. Cap volume per source (e.g. last 90–180 days) unless the user asks for full history.
Tier A — Highest signal for “how they work and sound”
| Source | What to fetch | Why |
| Slack | search.messages: from:user, date range, in:#relevant-channels; thread URLs they participated in | Tone, decisions, pushback, on-call voice |
| Slack AI (Slackbot DM) | Targeted questions: e.g. “Summarize how [Name] argues for design decisions in threads about [topic]” | Fast synthesis over large Slack corpus |
| Jira | JQL: assignee, reporter, comment ~, component/team filters; descriptions, comments, status transitions | Work scope, prioritization, written precision |
| GHE | PRs authored, reviewed (/pulls, review comments API); issues filed | Code review voice, technical standards |
| Bitbucket Server | Same pattern as GHE when Bitbucket Server is the primary Git host | Same |
Tier B — Depth and standards
| Source | What to fetch | Why |
| Confluence | Pages created by or substantially edited by them (CQL / search); team runbooks they own | Long-form standards, architecture voice |
| Notion | Pages they authored or commented on | Long-form async thinking, project context |
| SharePoint | Docs and wikis they own or edited | Standards docs, team handbooks |
Tier C — Optional / role-specific
| Source | When |
| Google Drive | Docs/slides they own (if verified in verified_connections.md) |
| PagerDuty | Oncall/incident behavior |
| Console / IAHub | Release/ops ownership if building an ops-heavy persona |
| Microsoft Teams / Outlook | If verified — email/thread tone (handle consent carefully) |
| Gmail (personal recipe) | Only if user explicitly wants email and connection is verified |
Tier D — Do not rely on for persona without extra care
- Raw git blame without PR context — noisy.
- HR systems — use only for title/team if needed, not personality inference.
Phase 3 — Synthesize (work vs persona)
Work (work.md): Systems, stacks, coding/review conventions, doc habits, Jira/workflow patterns, incident/release behavior — cite patterns, not one-off jokes.
Persona (persona.md): Use a layered structure compatible with colleague-skill:
- Layer 0 — Hard rules (non-negotiables: respect, no slurs, no real harassment simulation beyond professional friction the user explicitly asked for).
- Identity — role, scope, team context.
- Expression — vocabulary, sentence length, directness, humor.
- Decisions — risk posture, escalation, “how they say no.”
- Interpersonal — meetings, async, conflict.
Grounding: Prefer quoted paraphrases with source pointers; flag low-confidence traits when sample size is small.
Phase 4 — Write artifacts
meta.json (minimal)
{
"name": "Display Name",
"slug": "michael_donnelly",
"created_at": "<ISO8601 UTC>",
"updated_at": "<ISO8601 UTC>",
"version": "v1",
"profile": {
"company": "",
"level": "",
"role": "",
"email": ""
},
"ids": {
"slack_user_id": "",
"jira_account_id": ""
},
"knowledge_sources": ["slack", "jira", "ghe"],
"corrections_count": 0
}SKILL.md (invocable)
YAML frontmatter:
---
name: colleague_{slug}
description: "<Name> — distilled work + persona (tool-sourced)."
user-invocable: true
---Body: short intro + full work.md + full persona.md + run rules:
- Persona decides attitude; work block executes the task.
- Output matches persona expression.
- Layer 0 never violated.
Optional: also write work_skill.md / persona_skill.md with names colleague_{slug}_work / colleague_{slug}_persona if your host expects split invocations (see colleague-skill skill_writer.py).
Consent and safety
- Build skills only for legitimate work purposes and policy-compliant use of company tools.
- Do not exfiltrate secrets, PII bundles, or restricted content into
colleagues/— redact tokens, customer data, and health/financial identifiers. - When unsure whether content is allowed in a repo, ask the user before writing.
Outputs checklist
colleagues/{slug}/created withknowledge/raw/containing sourced excerptswork.mdandpersona.mdcompletemeta.jsonwith ids and source listSKILL.mdmerged and invocable- User told how to invoke (
/{slug}or host-specific command) and how to disambiguate duplicates (_2,_3)
Related skills
team_learner— team/domain bootstrap (similar tool sweep, different output shape).skill_creation— promote reusable methodology after validation.jira,tool_connections— all authenticated access.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

