agentsclimarketplace

Outreach

Skill manogyaguragai/recon-skill/skills/outreach

Write tailored cold emails to professors, admissions staff, recruiters, and programme coordinators for funded masters and PhD positions - and manage follow-ups and replies. Use when someone wants to email a professor, contact a supervisor, reach out about a research position or scholarship, needs help wording an academic cold email, wants their draft reviewed, asks how to follow up after no reply, or has received a reply and needs to respond well. Never sends anything; always hands the draft to the human.From its SKILL.md

Install
npx -y skills add manogyaguragai/recon-skill --skill outreach

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 11 stars11 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.7 KB, as published. Nobody here has run it

recon: outreach

Read ../recon/references/voice.md before drafting. It is the operative file for this stage.

Preconditions

Do not draft without a writing_sample in the inventory. Ask for one - any 200 words the user actually wrote, an old statement of purpose, a project README, a long message they sent someone. Two minutes of their time, and it is the single highest-leverage input to this stage. Drafting in a generic house voice is how this fails.

Do not draft without a verified hook. If target could not produce one that passes the find-and-replace test, say so and recommend either deeper reading or dropping the contact. A weak email is worse than no email - it does not fail neutrally, it burns the contact permanently and invisibly.

Check published_filters and obey them exactly.

Tracking setup - ask here, not at intake

The first time a user reaches this stage, offer the choice once:

Want me to put drafts straight into your Gmail drafts folder and watch for replies? Or I can hand you the text and you tell me when it goes out.

Record the answer in settings and never ask again. Both paths are fully supported and the manual one is the default that always works. Do not nag.

  • Manual - recon drafts, user sends from anywhere, user reports sends and replies. Zero dependencies
  • Gmail connected - drafts land in their drafts folder; thread search detects replies and computes follow-up windows
  • MCP - same, wired through a configured server in Claude Code

Recon must never require a connector. Most installs will not have one.

Recon never sends

Draft, hand over, human sends. This holds even where a send capability exists. The gap between drafting and sending is where a person reads their own email and catches the thing that would have embarrassed them. Removing that gap is how a tool becomes a spam machine.

Format by contact type

Academics - plain text, short, CV attached. No infographics, no video links, no branding. Visual production reads as marketing and marketing is precisely the smell being filtered for.

Industry, recruiters, employer-funders - different register, more warmth, and visual assets are normal and welcome. If the user wants a video, recon can write the script, shot list, and storyboard for them to record; recon cannot record it.

Draft, audit, revise

Never ship a first draft. Run the loop in voice.md:

  1. Draft using the four moves - hook, bridge, the user, the ask
  2. Ask explicitly what makes it read as machine written, and answer honestly
  3. Revise against that answer
  4. Find-and-replace test - swap in another name and institution. If it still works, the hook is generic. Go back
  5. Provenance test - every claim about the user traces to document or verified_link. Anything stated gets confirmed with the user or cut

Then show the user the draft and say what was verified and what was not.

The ask

Small, clear, easy to say yes to. Never open by asking for money.

Where the user has their own portable funding, or has applied for it, say so early and plainly - in systems where the supervisor pays from a project budget, it removes the single biggest barrier to hiring them. Where they do not have it, do not lead with the request.

Follow-ups

One short polite reminder after seven to ten days. One or two maximum. Never more. Recon computes when the window opens and tells the user; the user decides. A third follow-up converts nobody and costs the contact.

Replies

When a reply arrives - detected or reported - flag it immediately, update outcome and status, and offer to draft the response. Keep the loop open for that contact until it resolves into an outcome.

Match their length and register. A two-line question gets a short answer, not an essay. Answer what was asked before adding anything. This moment is the highest-leverage point in the whole process and the most commonly fumbled - usually by being too eager, too long, or too slow.

Record everything in outreach, stamping stages.outreach.last_run with the current ISO 8601 timestamp (date and time, e.g. 2026-07-20T14:32:00Z) in the same write. It is what makes later analysis possible, with the honesty caveat that small samples do not support strong conclusions.

Keep looking

Skills are one crate of 610,677. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.