communication

Communication: analyze & sharpen how you speak and write

Coach a real artifact the user provides — message, email, transcript, talk draft, CV, or resume — against proven rhetoric and structure techniques, then propose a concrete rewrite. The through-line everywhere: minimize the listener's cognitive load. The best-communicated idea wins, not the best idea.

Read On Demand

Style presets, by author (references/authors/)

Workflow

1. CONTEXT  → audience (peer / client / exec / group), channel (chat / email / talk),
              goal (inform / persuade / decide). Ask only if not inferable from the artifact.
2. DIAGNOSE → score the artifact against the rubric below. Note what works AND what's missing.
3. REPORT   → findings, severity-tagged (Blocker / Warning / Suggestion). Be specific, quote lines.
4. REWRITE  → produce a concrete rewrite applying the relevant techniques. Not abstract advice.
              If the user asked for an author style preset (DHH / Uncle Bob / Jessitron / Martin
              Fowler), load the matching file under references/authors/ and apply its moves on
              top of the rewrite.

Resume / CV field

Use this path for a CV or resume. Make it easy for a recruiter and ATS to answer: is this person a credible fit for this role? Ask for the target role and job description only when they are not inferable.

1. Position — replace a generic title with the role sought; open with a short executive summary that

states relevant scope, domain, and strongest proof.

1. Prove impact — rewrite duties as outcomes: strong verb + what changed + measurable result. Keep

numbers honest; never invent metrics.

1. Select signals — retain only skills, keywords, certifications, and experience relevant to the role;

remove weak or repetitive entries.

1. Make it scannable — use clear headings, standard role names, concise bullets, and an ATS-readable

layout. A polished design cannot compensate for unclear positioning.

Diagnostic rubric

• Ethos/Logos/Pathos — is credibility established? is the logic clear and low-load? is there an

emotional anchor (story, image, stakes)? Default delivery order: ethos → logos → pathos.

• Structure — for senior/structured audiences, is the conclusion first (Pyramid)? or buried?
• Focus — does it talk about them and their problem, or about the speaker/the idea/the solution?
• Safety — does adopting the idea feel risky/irreversible? is the risk normalized and reduced?
• Attention — is there a hook, or is it monotone/predictable?
• Resume positioning — for CVs/resumes: does the title and executive summary make the target role and

fit immediately clear? Do bullets prove impact instead of merely listing responsibilities?

The 7+1 Quick Reference

Full treatment with worked examples: references/frameworks.md.

Always-On Guardrails

• Cold, analytical tone by default. DIAGNOSE and REPORT in plain factual terms — no optimistic

framing, no cheerleading, no softening a weak artifact to make it feel better.

• Name risks and gaps explicitly. The REPORT step must state what's missing or fragile, not just

what works — a diagnosis that only lists strengths is incomplete.

• Rewrites are drafts, not final copy. The user reviews and refines every REWRITE himself — deliver

a clear, workable draft rather than over-polished prose he'd have to strip back down.

• Never fabricate ethos. Credibility claims must be true — don't invent credentials, numbers, or

experience to make an argument land. That's a lie, not a technique.

• Center the other person. The most common fix is moving the spotlight off the speaker.
• Lead with the conclusion for senior audiences. Buried recommendations get cut off with "where

is this going?".

• Stories over bare numbers. A figure embedded in a story/comparison is remembered far better than

the figure alone.

• Don't trust the 7-38-55 rule. It describes a narrow lab case, not normal speech (see references).
• Blunt ≠ personal (DHH mode). Directness targets the idea, never the person. And never soften

a requested DHH rewrite back into consensus-speak (see references/authors/dhh.md).

• Critique the work, not the person (Uncle Bob mode). Ruthless on the artifact, courteous to

whoever shared it (see references/authors/uncle-bob.md).

Anti-patterns

Integrated Example

Before (Slack to leadership): "Hi — so I've been thinking a lot and I really feel like I personally need more budget. I've done a ton of work this quarter and I'd love to be able to do more, there are loads of things I want to try and I think it could be good. Can we talk?"

Diagnosis: conclusion buried (no ask), self-focused ("I/me" throughout), zero ethos, zero logos (no reason), no pathos/stakes, feels open-ended and risky.

After: "We should double the team's budget this quarter. (1) We're behind on the roadmap and need to catch up; (2) we're resource-constrained on the highest-impact work; (3) the upside is real. To de-risk: start with a one-month pilot — fully reversible if the numbers don't move. That's how we get on the podium by quarter-end and actually celebrate hitting the targets together. Details on each point whenever useful."

Conclusion first (Pyramid) → 3 proofs (logos) → reversible pilot (safety) → shared win (pathos).

External References

• Aristotle, Rhetoric — ethos, logos, pathos.
• Barbara Minto, The Pyramid Principle (McKinsey) — conclusion-first communication.
• Mehrabian 7-38-55 — widely misquoted; valid only for single-word emotional/contradiction cues.
• Stanford — narratives are recalled markedly better than isolated figures.
• Source video: Quentin Despas / Katana, "Comment réfléchir vite et parler intelligemment".

Benchmark

Scenario: .benchmarks/scenarios/communication-001-message-coaching.md

PASS (run 2026-06-25). Gains on all three models (sonnet +29). Skill reliably adds conclusion-first Pyramid structure, the de-risk/pilot move, and a concrete rewrite that baselines partly miss. Gate per .agents/skills/skill-optimizer/rules/release-gates.md.

Scenario: .benchmarks/scenarios/communication-002-reframe-tough-question.md · Run: 2026-08-31 · Log: .benchmarks/runs/2026-08-31/communication-002-reframe-tough-question.json

NEUTRAL (run 2026-08-31). All models 100% with and without — reframing a tough question is already default behavior on this task; communication-001 above carries the skill's evidence. Gate per .agents/skills/skill-optimizer/rules/release-gates.md.