edmundmiller

just-say-no-to-process-porn

Stop AI process-porn, ceremony, and fake progress so work finishes as real software. Use at session start when the user wants anti-ceremony defaults; mid-session when the agent is planning, todoling, scaffolding, or status-reporting instead of shipping; before claiming done; or when the user mentions process porn, ceremony, performative work, reward hacking, or endless foundation/MVP theater. Do not use for genuine multi-stakeholder process the user explicitly requested.

edmundmiller 79 6 Updated 3w ago
GitHub

Install

npx skillscat add edmundmiller/dotfiles/just-say-no-to-process-porn

Install via the SkillsCat registry.

About this skill

This skill prevents AI agents from engaging in performative work, such as excessive planning, scaffolding, or status reporting, that does not result in functional software. It ensures progress is measured by working code rather than administrative ceremony. Use it when an agent prioritizes documentation or task lists over shipping actual deliverables.

SKILL.md

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.

Categories