Random Coffee Best Fit Outreach

Offline random coffee skill for ranking opt-in people and preparing consent-first intro packets. It creates local reports only; any external communication st...

Security Scan

The skill's requirements, instructions, and single helper script are coherent with an offline, consent-first random-coffee matching workflow and do not request unrelated credentials or network installs.

Purpose & Capability

Name/description ask for offline ranking and packet generation; the only runtime requirement is Python and the bundled wrapper script simply invokes a local CLI from the repo. No unrelated credentials, binaries, or config paths are required.

Instruction Scope

SKILL.md confines actions to local CSV inputs, local reports, and manual operator handoff. Commands shown run the local CLI/module and test suite. The instructions explicitly forbid external communication from the public skill and call out consent rules.

Install Mechanism

There is no install spec—this is instruction-only plus a small launcher script. Nothing is downloaded or written during install by the skill itself.

Credentials

No environment variables, secrets, or external credentials are requested. The skill expects operator-supplied, consented participant data and documents privacy-preserving practices.

Persistence & Privilege

The skill is not forced-always and does not request persistent system-wide privileges or modify other skills. It merely runs local code when invoked.

Assessment

This skill appears to do what it claims: run an offline matching workflow using local, consented CSV data and produce local review packets. Before using it: (1) verify participant data is properly consented and stripped of private profile text/handles, (2) run the included tests (pytest) and inspect the repository's src/random_coffee_matcher package locally to confirm behavior, (3) keep generated packets local and perform any outreach manually per the runbook, and (4) run the tool in an isolated environment if you want extra assurance (no network access required).

Goal

Run a consent-first random coffee workflow from local participant data:

• normalize opt-in people into a small participant CSV
• rank best-fit 1:1 intro candidates by mutual utility
• draft first-touch and double opt-in intro text
• render an offline review packet for the operator
• keep external communication outside this public skill

Use This Skill When

• the user asks for random coffee, best-fit introductions, warm networking, or founder/operator matching
• the source data is already opt-in, consented, or intentionally provided by the operator
• an older chat-first matching project exists and should be adapted into a public-safe intro workflow
• Codex should produce a repeatable intro packet, not ad hoc social copy

Inputs

Use a CSV with these canonical columns:

person_id,display_name,role,organization,location,timezone,languages,domains,skills,offers,needs,preferred_channel,availability,consent_notes,do_not_match,notes

Read references/intake-schema.md when the user gives messy notes, a contact map, or community notes.

Workflow

1. Restate the cohort goal, target audience, consent boundary, and verification command.
2. Normalize participant data into the CSV schema. Use placeholder or consented data only.
3. Rank matches:
   • In a cloned repo: python3 -m random_coffee_matcher rank <people.csv> --format markdown --out <report.md>.
   • From this skill wrapper in the repo: python3 {baseDir}/scripts/random_coffee_matcher.py rank <people.csv> --format markdown --out <report.md>.
4. Review the top matches. Prefer pairs with clear mutual utility, language overlap, manageable timezone gaps, and complete consent notes.
5. Generate a reviewed packet for any selected pair:
   • python3 -m random_coffee_matcher packet <people.csv> <person-a-id> <person-b-id> --out <packet.md>.
6. Hand the packet to the operator. Any external communication happens outside this public skill.
7. Log the operator-recorded outcome: skipped, blocked, opted in, declined, scheduled, or closed.

External Communication Boundary

Read references/outreach-surface-runbook.md before using the packet outside the repo.

Rules:

• Use only operator-provided or consented participant data.
• Keep the generated packet local until the operator approves it.
• Do not include private notes, long copied profile text, or private conversations in public artifacts.
• Do not reveal names, handles, links, or detailed context until both sides opt in.
• If any platform, privacy, or account-control issue appears, stop this workflow and ask for human handling outside the skill.

Outreach Rules

• First touch asks whether the person wants to be considered. It should not reveal another person's identity.
• Double opt-in asks each side before sharing names, handles, links, or detailed context.
• Keep drafts short, concrete, and easy to decline.
• Avoid fake urgency, pressure, claims of personal familiarity, or unverifiable praise.
• If either person declines or does not reply after the agreed follow-up limit, close the case.

Verification

For the open-source repo, run:

python3 -m pytest -q
python3 -m random_coffee_matcher rank examples/participants.csv --format text
python3 scripts/check_clawhub_skill_surface.py

Before publishing, run the local public-surface audit available in the surrounding Codex workspace when present.