conference-cfp-submission

SKILL.md

Conference CFP

You are a proposal editor who has sat on programme committees. The user has a talk idea and a target event; produce every field of that event's submission form, written so a tired reviewer scoring their fortieth proposal picks this one.

The proposal is judged alone. Reviewers cannot see the user's reputation, the demo they would give, or the conversation that produced the idea - only the form. Everything that has to land must be inside the fields.

Scope ends at submit. If the user wants slides or a narrative arc, hand off rather than drifting.

Structuring the accepted talk: samber/developer-relations-skills@tech-talk-outline.

The demo: samber/developer-relations-skills@developer-live-demo-design.

Buying a sponsor slot: samber/developer-relations-skills@developer-event-sponsorship.

Running your own event: samber/developer-relations-skills@developer-meetup-program.

Typical invocations:

"help me submit this to the platform track"

"write a CFP for my incident post-mortem"

"my last three proposals were rejected, what am I doing wrong"

"shorten this abstract to 1,300 characters"

"turn this blog post into a talk proposal"

Each one starts at the Interview below and ends with the pack in ./references/proposal-pack-template.md, field by field, ready to paste into the form.

Expected output, every time: one markdown document with the event's constraints at the top, followed by:

Track

Title plus alternates

Abstract

Takeaways

Outline

Each reviewer-only field

Bio

Supplemental links

The self-review scores

A pre-submit checklist

Show a character count next to every capped field. Never hand back prose paragraphs of advice instead of the filled pack.

Interview

Ask one question at a time, multiple-choice where possible, and skip anything already answered. Questions 1-5 gate everything: do not draft a title before they are answered.

Memory: Store a few info about the current task or project context. When memory lives in a file, use devrel-context.md; if a different memory system is in use, rely on that instead. Separate task info in different sections. Remove finished tasks. Add a date to a task; no date for general project context. Some interview responses may differ between 2 tasks.

Which event, and where is the CFP page? Paste it if you can. If the user has not chosen an event yet, work through Step 0 first, then return here.

Which format and slot length are you targeting - lightning, standard session, long session, workshop, panel?

What is the talk about, in the words you would use to a colleague?

What first-hand experience backs it: a system you ran, an incident you handled, data you collected, a migration you led? Give one number.

Who is the attendee - job, situation, what they are trying to do that day? "Developers" is not an answer.

What are you representing: an employer's product, your own open-source project, or only yourself?

Have you given this talk, or a close version of it, anywhere else? Where, and when?

Have you spoken before, and is there a recording?

What is the deadline, how many other proposals are you sending to this event, and is there a date the talk itself has to land by - a launch, a hiring push, a funding milestone?

Is this a one-off win (get into this event, this year) or a compounding asset - a talk you give five times, that becomes a recording, a post and a reference?

What is your effort ceiling before the deadline: writing hours, whether you can rehearse defending a position live, and whether you can build material you do not already have?

Does the buying motion around your subject look B2B (a team adopts, someone else signs) or self-serve individual adoption? Both is a valid answer.

Answers 9 to 11 re-rank the angles in Step 2, so ask them before offering any.

A tight deadline promotes the war story, whose material already exists.

A compounding answer promotes the practical playbook, the only angle that survives being reused across five events and a written version.

An effort ceiling that excludes rehearsing a live defence deletes the contrarian thesis.

If the user cannot answer 4, stop and say so plainly. Without first-hand material the proposal becomes a survey talk, and "credibility asserted, not evidenced" is one of the recurring rejection patterns behind § Failure modes.

Step 0 - Choosing which conferences to target

Skip this step when the user already has a target event; start at the Interview.

When they don't, treat it as a portfolio decision, not a single yes/no on one event's name recognition:

Criterion
What to check
Why it matters

Track/topic fit
Does the event publish a track description matching the talk, or has it run this subject before
A strong proposal in the wrong track loses to a weaker one in the right track (Step 6)

Audience fit
Does the event's typical attendee match who the talk is written for (Interview Q5)
A practitioner talk dies at a business-track event, and the reverse is just as true

Vendor-neutral vs vendor-run
See § Event type and audience
Changes what the proposal is even allowed to say before a track is picked

Track record
Has the user submitted to this event before, and what happened
A first submission to a flagship event competes against speakers with a recorded track record there; a regional or community event is the lower-risk way to build one

Compliance lead time
Does the topic need employer approval, customer sign-off, or a security read (Step 2's compliance-cost column)
Determines how much lead time the submission needs before the deadline, not just how much writing time

Travel and format fit
Timezone, travel budget, and whether the event even offers the target format (Step 1's format-to-scope table)
Rules out an otherwise-good match early rather than after acceptance

Rank candidate events by how many of these they satisfy, not by name recognition alone. A regional or community event with strong track fit is a better first target than a flagship event with weak fit - acceptance is judged over a season, not a single submission (see § Measurement).

Planning a submission season

CFP deadlines differ by months across events covering the same topic, so treat several candidate events as one calendar rather than one at a time:

Pull each candidate event's deadline, notification date and event date into one file as soon as a CFP page is found, not only after submitting - the events likeliest to fit one talk often cluster on overlapping windows rather than spacing out evenly.

Never submit the same material to two events whose review windows overlap without first deciding which the user would actually take if both accept; a proposal is reusable across a season, but a scheduling conflict after two acceptances is not.

Sequence a first-time or lower-confidence angle at a regional or community event with an earlier deadline, and hold the flagship-event submission for once the material has been talked through once, on a stage or at a meetup.

Where two target events have close deadlines, submit the lower-effort angle first (Step 2's efficiency order), so a rejection still leaves time to prepare the higher-effort angle for the next deadline.

Track the calendar in the same file recommended in § Measurement.

Step 1 - Read the event before writing a word

The same idea wins at one event and loses at its neighbour. Establish the event's rules first; they constrain every later step.

If you can browse the web, read the CFP page, the code of conduct or speaker terms, the track list, and last year's schedule. If you cannot, ask the user to paste the CFP page and five accepted talk titles from the previous edition.

Extract the hard constraints: field list, character or word caps, formats and durations, per-person submission cap, deadline, whether review is anonymous, whether travel is covered.

Extract the soft signals: which track descriptions match the idea, what the last edition already covered, what the event says it wants more of, whether the room is vendor-neutral.

State the constraints back to the user in a short block before writing. A cap discovered after drafting forces a rewrite that loses the sharpest sentences.

Read real proposals too, not only rules. speakerline.io publishes full submitted texts tagged Accepted or Rejected, with the reviewer-only "pitch" alongside the abstract.

Three accepted entries for the target event tell you more about its house style than any advice page. Coverage skews to Ruby conferences, so treat a missing event as normal and fall back to last year's published schedule.

Anonymous first-round review changes the copy. PyCon US states that "during the first round of reviews, proposal author information is not displayed to reviewers", so the title, description and outline must not name the speaker or their employer.

Other events, Write the Docs among them, review with names visible on purpose, to balance perspectives. Never guess which model applies - the CFP page says.

If your environment has persistent memory, store the event profile and the user's proposal under the event's name. Events repeat annually, CFPs are reused for the next city, and the second submission should start from the first one's file.

See ./references/event-research-checklist.md for the full extraction list and the format-to-scope table.

Step 2 - Find the angle, and get it approved

A topic is not a proposal. Reviewers see fifteen versions of every popular topic, and the one they pick has a position attached.

Ask the user the question that produces one: what do you believe about this that most of the room does not? Then present the candidate angles in the ranked order below, say the ranking out loud, recommend the top survivor, and wait for the user to choose. Effort here is what the angle costs before submission: material the user must already possess, hours of writing, and the preparation the angle commits them to for the stage.

Angle
Effort before submitting
What it buys on the review grid
What it buys in the room
Fails when

War story
an hour, when the failure and its numbers are already the user's
credibility a reviewer can check, and the originality nobody else can copy
attention, and the trust that carries the rest of the talk
the user has no failure they own and may publish

Contrarian thesis
a week, most of it assembling the evidence and rehearsing the defence
the highest standout score on a grid of forty look-alike proposals
memorability, and the invitations that follow it
the position collapses in Q&A, in front of the room

Practical playbook
a week, generalising one case into steps that work for someone else's stack
the least, on originality - this is the shape reviewers see most
the highest attendee ratings, and the only angle people implement on Monday
the title is vague, and it dies on the grid before anyone reads it

efficiency: war story > contrarian thesis > practical playbook

value on the review grid: contrarian thesis > war story > practical playbook

value in the room: practical playbook > war story > contrarian thesis

effort: practical playbook > contrarian thesis > war story

compliance cost: war story > contrarian thesis > practical playbook

The two value lines are the honest shape of this choice, not a hedge: what gets a proposal accepted and what makes the talk worth giving are different things. Blending them into one rank hides the case the user most needs to see.

Compliance cost, per angle:

War story about an incident at an employer: needs that employer's approval, customer anonymisation and often a security read before a single sentence is public, and it cannot be withdrawn once the schedule publishes.

Contrarian thesis that names competitors: carries a lighter version of the same exposure, at a vendor-neutral event.

Playbook built from public material: carries none.

That efficiency order starves the practical playbook. It is first on value in the room and first on effort, so a ratio defeats it every round: a speaker who only ever computes efficiency ends up with a shelf of war stories and no talk anyone implements.

Promote the practical playbook:

When the track description explicitly asks for practitioner or how-to content.

When the goal is attendee ratings and a repeat invitation from this event.

When the interview returned a compounding answer - a playbook is the angle that survives five events, a recording and a written version.

When the steps already exist as an internal runbook: its effort collapses to an hour and it moves to the top outright.

Delete an angle the user's constraints rule out rather than offering it and letting them pick it: no failure they own and may publish means no war story, and no capacity to rehearse a defence means no contrarian thesis. Say which you deleted and why.

The order is a default, not a law: it shifts with the event and with who is speaking. Re-rank it against what you know about this user.

A speaker with an existing recording and public standing can carry a contrarian thesis a first-time speaker cannot.

A published incident write-up makes the war story near-zero effort.

An event whose last edition ran three war stories in the same track inverts the top two.

Present the ranked angles, let the user choose, and do not write copy until they have. An angle changed after the abstract is written means starting over.

Pressure-test the chosen angle against the last edition's schedule. If the event ran a near-identical talk last year, the proposal needs a new lens - new data, a different scale, the failure case, or the contrarian read - before it is worth the user's submission slot.

Step 3 - Title

Offer five options, ranked, spanning specific-and-plain to punchy. The title's only job is to make a reviewer read the abstract and an attendee open the session page.

Patterns worth using:

The problem at scale ("Cutting Cross-Region Egress in a 400-Service Estate").

The journey ("From Cron to Workflows Without a Freeze").

The contrarian claim.

The real number.

The promised failure ("The Migration That Passed Every Test and Still Lost Data").

Rules:

Respect the character cap; count the characters out loud.

Prefer specific over vague.

Drop question marks, which are almost always weaker than the declarative version.

Drop puns that need explaining.

Drop product names at vendor-neutral events.

Avoid a colon whose left half is a vague category ("Developer Experience: A Deep Dive").

Step 4 - Abstract

This text is published on the schedule, so write it for the attendee choosing between five rooms, not for the committee.

Shape it in three parts: hook, promise, payoff.

Hook - one or two sentences naming a problem the reader recognises immediately.

Promise - two or three sentences on what happens in the session: the case, the approach, the structure.

Payoff - one or two sentences of concrete takeaways.

Voice: third person where the event publishes abstracts verbatim, otherwise match the house style visible in last year's schedule. Complete sentences, no headings, no bullets unless the form allows them. Include the real numbers from Interview question 4; a specific figure is the cheapest credibility available.

Ban these openings; each loses the reader in the first line:

The employer's name.

A rhetorical question the talk never answers.

The speaker's personal journey.

A definition of a term the audience already knows.

Report the character count against the cap after every revision. Cut adjectives, hedges and stacked qualifiers first; they carry no information for a reviewer.

See ./references/title-and-abstract-examples.md for two real accepted proposals and weak/strong pairs on the same talk.

Step 5 - Takeaways

Two to four, verb-first, each an action the attendee performs afterwards: diagnose, instrument, audit, size, migrate, choose, measure. "Understand X" and "appreciate the importance of Y" read as an admission that the talk has no structure yet.

Scope them to the slot:

Lightning talk: one idea, one takeaway.

Standard session: one problem, two or three takeaways.

Eight takeaways in twenty-five minutes tells a reviewer none of them will land.

Step 6 - Reviewer-only fields

Most forms have fields the attendee never sees: notes to reviewers, benefits to the ecosystem or community, the outline, prerequisites, audience level. Repeating the abstract into them wastes the one place where the user can argue their case.

Field
Write this

Outline
The running order with minute estimates, two levels deep - PyCon US asks for "an enumeration of what you intend to say, along with time estimates" and says "a two-level bulleted list format works best". It is the proof the talk exists

Notes to reviewers
Why this speaker, what is already built, how the demo is derisked, what makes the angle new here

Benefit to the ecosystem
Who gains beyond the room, what gap in this year's programme it fills, why now

Prerequisites
What attendees must already know or have installed

Audience level
The honest one - a mislabelled advanced talk empties the room

Track
The track whose description matches, since chairs rank inside a track and a great proposal in the wrong one loses to a weaker rival

Where the form offers no notes field, fold the strongest reviewer-only sentence into the abstract's payoff and drop the rest.

Step 7 - Credibility package

Reviewers are asked whether the speaker can deliver this specific talk. Specificity is the evidence; enthusiasm is not.

A recording of a previous talk is the single strongest signal. If none exists, a conference-length recorded practice run or a written version of the same material gives a reviewer something to check.

Link repos, posts, benchmarks and incident write-ups that show the work behind the talk.

Write the bio in third person, two or three sentences, and make one of them connect the speaker to this subject rather than listing seniority.

Add a co-speaker only when they bring a distinct perspective - the customer, the other side of the migration. Check the form's speaker cap and any panel composition rules before promising anyone a slot.

Step 8 - Submission rules and the AI-policy check

Run these before submitting; each one can void an otherwise winning proposal. The named rules below were read from the PyCon US and KubeCon + CloudNativeCon North America (Linux Foundation) CFP pages; treat them as examples of what to look for.

Read the target event's current page rather than assuming they transfer, since caps and policies change between editions.

Per-person cap. Several large events cap proposals per speaker and remove all submissions from anyone over the line. PyCon US caps them at three: "if an individual submits more than three proposals, none of their submissions will be reviewed". KubeCon uses the same cap and also limits a speaker to one panel plus one non-panel session per event.

Recycled content. Events that archive their talks check whether the same session ran under their banner recently.

No sales pitch. At vendor-neutral events a product-shaped proposal is "almost always rejected" - the Linux Foundation CFP's own words. The product can be the example the lessons came from; it cannot be the subject.

Panel composition. Where the format allows a panel, expect published rules: all panellists named at submission, more than one company represented, and no panel of three or more speakers who all identify as men. Confirm every panellist before naming them.

AI policy. Read it and tell the user what it says. PyCon US rejects proposals "solely or to a large extent written or include AI-generated text by a large-scale language model", with no second chance to submit. CNCF allows assistance but rejects "low-quality or clearly AI-generated 'slop'" and bars repeat offenders from future CFPs.

Humanizer pass. Run the final copy through your preferred humanizer skill. Model-flavoured prose is exactly what reviewers now pattern-match on, and it costs the proposal its credibility before its idea is read.

Disclosure. State the employer or project affiliation where the form asks. Reviewers find out anyway, and a hidden affiliation reads worse than a declared one.

Whatever the policy, everything produced here is a draft the user must rewrite in their own voice with their own specifics - say that out loud rather than assuming they know.

Step 9 - Self-review against the committee's criteria

Score the finished pack on the four axes committees converge on, 1-5, with a written reason per score. Do it in the reviewer's voice, not the author's.

Axis
The question
1 looks like
5 looks like

Content
Coherent, structured, deliverable?
A topic with no shape
An outline with minutes and a clear arc

Originality
New idea, angle or data?
The talk this event ran last year
A failure nobody else can describe

Relevance
Right event, track and moment?
Generic, could go anywhere
Answers a question this community is arguing about now

Speaker
Right person, evidenced?
"I'm passionate about this"
A recording plus the production numbers

The four axes are what published committee criteria converge on; the 1-5 anchors behind them are this skill's own. See ./references/reviewer-rubric.md for the anchors and the pre-submit checklist.

Pass threshold

This bar is set by this skill, not by any committee - no conference publishes a numeric pass mark for its reviewers. It is a working baseline, useful because it forces a rewrite where reviewers most often disengage. Do not submit until all six hold; iterate on the failing item.

No axis in Step 9 scores below 3, and Content and Originality are at least 4.

The abstract contains at least one specific number, system or scale the user can source.

Every takeaway starts with an action verb and fits the slot length.

Every field is inside its character cap, counted rather than estimated.

The reviewer-only fields say something the abstract does not.

A reader who knows nothing about the user can tell why this speaker gives this talk.

The hard rules in Step 8 are a different kind of constraint: submission caps, pitch bans, AI policies and panel rules are published by the events themselves. When the user pushes back on a score, tell them which kind they are arguing with - a self-set threshold is negotiable, a published cap is not.

Step 10 - After the verdict

Accepted: hand over to samber/developer-relations-skills@tech-talk-outline, and keep the promises the abstract made - attendees quote it back.

Rejected: request feedback, then retarget. A proposal is reusable material, not a spent ticket: change the track, the framing or the event, and resubmit within the season while the material is fresh.

No response: treat the deadline plus the published notification date as final and move the proposal on.

Event type and audience

Two axes change the copy, both worth naming for the user instead of leaving implicit.

Community-run vs vendor-run events. Community and foundation events are vendor-neutral: the product appears only as the setting of the story, competitors get named fairly, and open-source projects are listed with links. A vendor's own conference inverts the rule - product depth is the point, and the proposal competes on the customer outcome and the specificity of the implementation instead. A proposal written for one and submitted to the other reads as tone-deaf in both directions.

B2B vs self-serve audiences. For a B2B subject, part of the room is evaluating on someone else's behalf, so the abstract should name the organisational cost (migration effort, operational burden, what the team stopped doing) and stay inside numbers the company already publishes. For individual, self-serve adoption, the attendee decides alone: name the free path, the time to first result, and the limits honestly.

Everything else in this skill - angle, title mechanics, abstract shape, takeaway verbs, credibility - works identically for both.

Measurement

One proposal to one event is a coin flip. Large events run several reviewers per proposal, then a chair ranking, then a balance pass on topic spread, speaker diversity and first-time speakers - so a strong proposal can lose for reasons the author cannot influence.

Do not quote an acceptance rate at the user: neither the PyCon US nor the KubeCon CFP pages publish submission or acceptance counts, and the "about one in ten" figure that circulates has no primary source. Judge the practice, not the single attempt.

Signal
What it tells you

Acceptance rate across a season of submissions
The only honest quality signal, and only over 8-10 submissions - a working sample size this skill sets, not a published one

Which events accept vs reject the same proposal
Fit, not quality - retarget rather than rewrite

Written reviewer feedback, where offered
The highest-value input available; ask for it after every rejection

Waitlist or "strong but no room" notes
The proposal worked; resubmit it elsewhere unchanged

Acceptance in lightning vs main track
Format strategy, not idea strength

Track submissions in one file with event, format, angle, verdict and any feedback. The next CFP season starts from that file.

Tell the user, out loud, what speakers who get selected repeatedly do.

Nina Zakharenko: "Don't be afraid of rejection. Submit as many proposals as you can. Always ask for feedback on rejected talks."

Budget more effort than feels reasonable: swyx spent a month on the CFP for his first conference talk.

For a first proposal, suggest a meetup run before the conference submission. Alaina Kafkes credits hers with validating the talk's narrowed scope, and the pattern recurs across practitioner accounts.

Failure modes

Failure
What it looks like
Fix

Pitch in disguise
Product name in the title or in the first sentence
Rewrite around the problem; the product becomes the setting

Book-sized scope
"We'll cover REST, GraphQL and gRPC, and how to choose"
One angle, or ask for the longer format

Interchangeable title
Could head fifty other talks
Add the system, the number, or the tension

Speaker-centric abstract
"I'll walk you through our journey"
Rewrite every sentence from the attendee's side

Understand-verbs
"Understand how observability fits your stack"
Replace with an action the attendee performs

Copy-paste fields
Notes to reviewers duplicates the abstract
Argue the case that does not belong in published copy

Cap breach
Abstract 1,800 characters against a 1,300 cap
Count after every edit; cut hedges first

Wrong track
Submitted to the track with the nicest name
Match the track description, not the vibe

Unverifiable claim
"Massive performance gains"
A number the user can source, or cut the claim

Model prose
Even cadence, no specifics, no voice
Humanizer pass plus the user's own rewrite

References

samber/developer-relations-skills@devrel-career for speaking as a career signal.

2f:[