imMamdouhaboammar/anti-slop-engineering / skills/cognitive-load-and-attention-governor
cognitive-load-and-attention-governor
VerifiedEnforce cognitive bandwidth and attention economy invariants across AI agent outputs, human-in-the-loop interactions, pull request descriptions, standalone deliverables, and multi-agent swarm handoffs. Governs scannability metrics (time-to-point, max wall length, landing anchor density), terminal blocking-question placement, deliverable purity (zero opener/closer fluff), working memory list bounding (Miller's law <= 5), and non-regression constraint preservation.
Cognitive Bandwidth & Attention Economy Governor
"Attention is the scarcest resource in this conversation, and you are spending it with every word."
The cognitive bandwidth invariant execution pipeline is:
$$\text{Raw Content} \xrightarrow{\text{Scannability Gate}} \mathcal{S} \xrightarrow{\text{Terminal Ask Check}} \mathcal{B}{pos} \xrightarrow{\text{Deliverable Purity}} \mathcal{P}{deliv} \xrightarrow{\text{Non-Regression Guard}} \text{Pure Payload}$$
Why This Skill Is Needed & Target State
The Problem: Most LLM agents suffer from communicative slop:
- The Buried Ask: When an agent needs human confirmation before an irreversible action (e.g. database migration, git push, deleting files), it buries the question in paragraph 2 and follows it with three paragraphs of status updates. Busy or ADHD users skim past it and miss the pending decision.
- Deliverable Impurity: When asked to write a commit message, PR description, or configuration block, agents wrap it in conversational framing ("Here is your commit message:", "I hope this helps! Let me know if you need anything else!"). This breaks automated CI pipelines and wastes reviewer focus.
- Impenetrable Prose Walls: Text that scores well on traditional readability formulas (Flesch-Kincaid) often consists of dense, anchor-free paragraphs. Readers bounce off walls of text without absorbing the load-bearing facts.
- Agent-to-Agent Swarm Bloat: In multi-agent systems (Fable, Jev, EGM goldfish reviews), verbose subagent status reports contaminate orchestrator context windows, leading to attention decay and missed failure conditions.
Target State: Zero conversational wrappers on deliverables. Bottom-line on Line 1. High visual anchor density ($\ge 3.0$ anchors per 100 words). Maximum unbroken wall $\le 60$ words. All blocking questions positioned in the terminal block with Line 1 flagged. Working memory lists capped at 5 items. Zero dropped technical numbers, thresholds, or warnings.
Core Invariants
1. The Terminal Blocking-Question Invariant ($\mathcal{B}_{pos}$)
- If the reply requires human confirmation or a decision before proceeding, the ask MUST sit in the last non-empty block, with zero substantive prose after it (only short options/yes-no affordances permitted).
- If the reply exceeds 40 words, Line 1 MUST explicitly flag the ask (e.g.
before I,need,confirm,sign-off,go-ahead,wait, or a decisive verdict openerNo,Don't,Stop,Hold).
2. Deliverable Purity Invariant ($\mathcal{P}_{deliv}$)
- When asked to produce a deliverable (code, commit message, PR body, README snippet, config, command):
- Zero Opener Wrappers: Ban
Here is...,Below is...,Sure!,I've drafted... - Zero Closer Offers: Ban
Let me know if...,Hope this helps!,Feel free to customize... - Zero Lead-in Preamble: Ban transitional lines ending in
:or.right before a fenced code or quote block.
- Zero Opener Wrappers: Ban
3. Text Shape Scannability Invariant ($\mathcal{S}$)
- Time-to-Point ($TTP \le 20$ words): Distance to the first bold anchor or verdict.
- Maximum Wall Length ($W_{max} \le 60$ words): Longest paragraph without a bold word, bullet, or heading.
- Anchor Density ($\rho_{anchor} \ge 3.0$ per 100w): Visual anchors (
**...**,β,#,-) every 30 words. - Answer-First ($\mathcal{A}_1 = 1$): Non-empty Line 1 starts with a bold takeaway or direct verdict.
4. Working Memory & List Bounding ($\mathcal{M}_{cap}$)
- Flat lists must not exceed 5 items (Miller's Law).
- For broader surveys, enforce progressive disclosure: name the top 2-3 items in full, and name/offer the remainder.
5. The Non-Regression Invariant (Preserving Technical Rigor)
- Brevity must never drop:
- Exact numbers, timeouts, durations, and byte limits.
- Scoped conditions (e.g., "only workspaces < 14 days old").
- Safety warnings, error codes, and operational caveats.
- A warning is the last word to cut, never the first.
Execution Workflows
| Mode | Trigger Condition | Playbook | Output Style |
|---|---|---|---|
| Attention-Kind | Default interactive agent dialogue | playbooks/human-agent-interaction.md#attention-kind | Plain English, front-loaded, bold anchors |
| Spartan | Heads-down coding, critical commands | playbooks/human-agent-interaction.md#spartan | Blunt, imperative, arrow bullets (β), zero warmth |
| Rundown | Standups, phase reviews, status checks | playbooks/human-agent-interaction.md#rundown | TL;DR on line 1, checklist (β
,π‘,β¬,β), Your move: |
| Swarm TOON | Subagent-to-orchestrator handoffs | playbooks/agent-to-agent-swarm.md | Compressed TOON payload, bounded state tokens |
Mathematical Metrics & Reference
Detailed mathematical definitions, KaTeX formulas, and empirical verification bounds live in references/attention-metrics.md.
Formal inter-module handoff contracts with Fable, J-Space, OmniSkill, OpenReview, PR Boundary Guard, and Comment Budget Guard live in references/handoff-contracts.md.
CLI & Verification Commands
# Audit markdown file for scannability and attention violations
bunx anti-slop-engineering attention --file docs/rfc.md --json
# Verify 100% deliverable purity on a generated artifact
bunx anti-slop-engineering attention --text "$COMMIT_MSG" --check-deliverable
# Audit git diff additions across documentation
bunx anti-slop-engineering attention --base HEAD
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why β repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" β or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself β your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |