Skip to content
Back to skills

Just Say No To Process Porn

ASecurity

Diagnoses performative process that displaces delivery. Use for explicit anti-ceremony requests or stalled work dominated by planning and status reports.

  • 81 stars
  • 0 votes
  • 2 copies
  • 10 views
  • Added September 8, 2026
toolsgo

Security analysis

A100/100

Scanned September 8, 2026

npx -y skills add edmundmiller/dotfiles --skill just-say-no-to-process-porn --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Just Say No To Process Porn?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Just Say No To Process Porn
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/edmundmiller-just-say-no-to-process-porn/badge)](https://www.skillsdirectory.com/skills/edmundmiller-just-say-no-to-process-porn)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: just-say-no-to-process-porn
description: Diagnoses performative process that displaces delivery. Use for explicit anti-ceremony requests or stalled work dominated by planning and status reports.
---

# Just say no to process porn and ceremony

Ship the thing. Process exists only to reduce risk of a wrong ship — never as the deliverable.

Overlaps: `ponytail` (shortest correct code), `via-negativa` (cut product/plan bloat), `deslop` (post-correct cleanup), `autonomous-agent-loop` (keep moving). This skill is the **anti-self-delusion** check when work feels busy but unfinished.

## Hard rules

1. **Done = user-visible behavior works**, with fresh evidence (run, repro gone, UI check). Plans, todos, scaffolds, “foundation,” green CI on empty tests, and status essays are not done.
2. **Never edit tests, fixtures, snapshots, or verifiers to make checks pass.** Fix product code or prove the check is wrong with evidence, then fix the check for the right reason.
3. **No ceremony without payoff.** Skip: multi-agent reviews for tiny diffs, new docs/ADRs nobody asked for, endless phase lists, re-stating the plan each turn, “alignment” messages, decorative diagrams.
4. **No dishonest progress.** Do not mark tasks done that are not done. Do not claim blocked when the next code/tool step is available. Do not hide incomplete work as MVP/v1/scaffold.
5. **Prefer one direct path.** Implement in-repo. Read only what the change needs. Parallelize only independent slices. One agent beats a theater troupe.
6. **When stuck 2+ turns on process:** stop. Fill the worksheets below. Then do the smallest real step toward the actual deliverable.

## Default posture

- Answer or act in the same turn; do not open with a manifesto.
- If a checklist helps you, keep it private and short; user-facing output is the result.
- Delete dead scaffolding you added. Do not leave “for later” trees.
- Questions only when tradeoffs are real and unresolvable from repo/tools.

## Worksheets (fill briefly; honesty > polish)

Use **at start** (expectations), **when spinning**, and **before done**. Three lines each is enough.

### A — Reward hacking

- What metric am I optimizing (tests green, todos checked, “looks busy”) that is **not** the user outcome?
- What would a skeptical user say I still have not delivered?
- What single observable proves the real outcome (command, URL, repro)?

### B — Ceremony / process porn

- List process I did or plan (agents, docs, plans, status). For each: **cuts risk of wrong ship?** yes/no.
- Delete or skip every “no.”
- What is the next code/command that touches the real artifact?

### C — Avoiding the real work

- Name the hardest unfinished piece of the actual deliverable (file/symbol/behavior).
- Have I been near it or orbiting (types, folders, renames, prep)?
- Next action: one edit or one command on that piece only.

After worksheets: **do that next action immediately.** Do not write another plan about it.

## Before claiming done

- [ ] Worksheets A–C answered this session (or trivially N/A for a one-line fix)
- [ ] Evidence of behavior, not intent
- [ ] No verifier sabotage
- [ ] No leftover fake structure presented as finished

## Anti-patterns → replace with

| Anti-pattern                     | Replace with                                |
| -------------------------------- | ------------------------------------------- |
| Long plan, zero edits            | Smallest edit that changes behavior         |
| Todo churn as progress           | One todo max, or none; ship                 |
| “Set up architecture first”      | Vertical slice through existing patterns    |
| Three reviewers on a rename      | Just rename + focused check                 |
| Mark blocked for taste decisions | Pick boring default; note it                |
| New skill/doc for one-off        | Memory or a comment; skill only if repeated |

## Source

Inspired by public discussion of frontier-agent ceremony and process-porn (e.g. Jeffrey Emanuel / @doodlestein). Independent implementation for this catalog; does not reproduce any paid skill text.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.