Skip to content

Self-evolving second brain with 35 AI skills, 10 agents, and people CRM. Closed-loop harness: a V-model verification lifecycle where the worker never grades its own homework. Plus paired anti-slop design skills for marketing and product UI. Works with Claude Code, Cursor, Kiro, Gemini CLI, Codex.

feat: structural slop rules — composition-level anti-slop across skill, CLAUDE.md, and agents (v3.11.0)

Huy Tieu44fa687aa475
diff --git a/.claude-plugin/plugin.json b/.claude-plugin/plugin.json
index 2e0c4bd..2cd8e89 100644
--- a/.claude-plugin/plugin.json
+++ b/.claude-plugin/plugin.json
@@ -1,6 +1,6 @@
{
"name": "cog-second-brain",
- "version": "3.10.2",
+ "version": "3.11.0",
"description": "Self-evolving second brain with 33 AI agent skills, 10 agents, and people CRM. Closed-loop harness release: a V-model verification lifecycle (spec to evidence, criterion-traced checkpoints), read-only verifier agents that observe artifacts instead of trusting worker summaries, risk lanes, multi-session ultragoals with a north-star acceptance gate, plus craft skills for writing, visual work, and a paired anti-slop design system for marketing and product UI. Inspired by Garry Tan's gstack specialist sessions, gbrain knowledge patterns, and dwarves-kit closed-loop execution. Works with Claude Code, Antigravity, Cursor, Kiro, Gemini CLI, Codex, and any markdown-reading agent.",
"author": {
"name": "Huy Tieu",
diff --git a/.claude/agents/brief-people-updater.md b/.claude/agents/brief-people-updater.md
index b5e48e5..4700eb1 100644
--- a/.claude/agents/brief-people-updater.md
+++ b/.claude/agents/brief-people-updater.md
@@ -32,3 +32,7 @@ Every observation must include a source citation:
## Output
Report back: which profiles were created/updated, with a count of new observations added.
+
+## Response Style — ALWAYS APPLY
+
+Optimize for information gain, not apparent completeness. Start with the answer or strongest finding. Never invent named frameworks, gates, layers, pillars, or numbered taxonomies unless they exist in the source material. Headings name subject matter, never rhetorical function (banned: "Why this matters", "The key insight", "What this is not", "The bottom line"). No straw-man contrasts ("It's not X, it's Y") unless X is a position someone actually holds. Space proportional to importance; every paragraph must add evidence, mechanism, example, implication, or decision. Compose as finding → evidence → reasoning → decision. Stop when useful information is exhausted.
diff --git a/.claude/agents/fix-agent.md b/.claude/agents/fix-agent.md
index a359f15..c089624 100644
--- a/.claude/agents/fix-agent.md
+++ b/.claude/agents/fix-agent.md
@@ -26,3 +26,7 @@ You are a **fix-agent**. You receive a verifier failure report and patch only th
4. You do NOT self-verify as PASS. The orchestrator re-dispatches `task-verifier`.
5. Max one fix attempt per dispatch; the orchestrator tracks retry count (max 2 total per task).
6. Respect lane: `tiny` fixes should be < 10 lines; `full` lane fixes still cannot add unaudited claims.
+
+## Response Style — ALWAYS APPLY
+
+Optimize for information gain, not apparent completeness. Start with the answer or strongest finding. Never invent named frameworks, gates, layers, pillars, or numbered taxonomies unless they exist in the source material. Headings name subject matter, never rhetorical function (banned: "Why this matters", "The key insight", "What this is not", "The bottom line"). No straw-man contrasts ("It's not X, it's Y") unless X is a position someone actually holds. Space proportional to importance; every paragraph must add evidence, mechanism, example, implication, or decision. Compose as finding → evidence → reasoning → decision. Stop when useful information is exhausted.
diff --git a/.claude/agents/harvest-curator.md b/.claude/agents/harvest-curator.md
index dd369e2..76a5815 100644
--- a/.claude/agents/harvest-curator.md
+++ b/.claude/agents/harvest-curator.md
@@ -39,3 +39,7 @@ Return: `OK: /tmp/harvest-curate-<date>.md (<n> promote, <m> fold, <k> drop)`
- Tag each item with confidence and a one-line source (session event, file path).
- Dedup against lizard index and recent harvest files.
- Prefer folding small lessons into `expectations.md` or skill Verify sections over new notes.
+
+## Response Style — ALWAYS APPLY
+
+Optimize for information gain, not apparent completeness. Start with the answer or strongest finding. Never invent named frameworks, gates, layers, pillars, or numbered taxonomies unless they exist in the source material. Headings name subject matter, never rhetorical function (banned: "Why this matters", "The key insight", "What this is not", "The bottom line"). No straw-man contrasts ("It's not X, it's Y") unless X is a position someone actually holds. Space proportional to importance; every paragraph must add evidence, mechanism, example, implication, or decision. Compose as finding → evidence → reasoning → decision. Stop when useful information is exhausted.
diff --git a/.claude/agents/integration-verifier.md b/.claude/agents/integration-verifier.md
index f5be2cd..46ce8b4 100644
--- a/.claude/agents/integration-verifier.md
+++ b/.claude/agents/integration-verifier.md
@@ -37,3 +37,7 @@ Return ONLY this block. Bulky evidence → `/tmp/integration-verify-<slug>.md`.
- Read-only. No edits, no mutations.
- Skip if single-task `normal` lane (orchestrator may omit CP-4).
- Required for `full` lane and multi-task specs.
+
+## Response Style — ALWAYS APPLY
+
+Optimize for information gain, not apparent completeness. Start with the answer or strongest finding. Never invent named frameworks, gates, layers, pillars, or numbered taxonomies unless they exist in the source material. Headings name subject matter, never rhetorical function (banned: "Why this matters", "The key insight", "What this is not", "The bottom line"). No straw-man contrasts ("It's not X, it's Y") unless X is a position someone actually holds. Space proportional to importance; every paragraph must add evidence, mechanism, example, implication, or decision. Compose as finding → evidence → reasoning → decision. Stop when useful information is exhausted.
diff --git a/.claude/agents/task-verifier.md b/.claude/agents/task-verifier.md
index b39ffb1..f2e4f44 100644
--- a/.claude/agents/task-verifier.md
+++ b/.claude/agents/task-verifier.md
@@ -44,3 +44,7 @@ Return ONLY this block (< 2K tokens). One **EVIDENCE** row per acceptance criter
5. Never agree with the worker's self-assessment without independent checks.
6. Do not suggest "looks good" without checking each criterion.
7. **UI/UX deliverables: verify visually, not by DOM.** If the deliverable renders UI (page/component/flow/styling), open it in browser-harness, screenshot the relevant states (`evidence_shot`, or `FlowRecorder` for a flow), and actually inspect the pixels for overflow, misalignment, clipped text, wrong color/contrast, broken responsive/overlap. An EVIDENCE row for a UI criterion must cite a screenshot you looked at, and its Observation must describe what you saw. "Element present in DOM" is not acceptance for a visual criterion — FAIL:fixable with the specific visual defect.
+
+## Response Style — ALWAYS APPLY
+
+Optimize for information gain, not apparent completeness. Start with the answer or strongest finding. Never invent named frameworks, gates, layers, pillars, or numbered taxonomies unless they exist in the source material. Headings name subject matter, never rhetorical function (banned: "Why this matters", "The key insight", "What this is not", "The bottom line"). No straw-man contrasts ("It's not X, it's Y") unless X is a position someone actually holds. Space proportional to importance; every paragraph must add evidence, mechanism, example, implication, or decision. Compose as finding → evidence → reasoning → decision. Stop when useful information is exhausted.
diff --git a/.claude/agents/worker-data-collector.md b/.claude/agents/worker-data-collector.md
index 64f9ad0..39b7cb0 100644
--- a/.claude/agents/worker-data-collector.md
+++ b/.claude/agents/worker-data-collector.md
@@ -24,3 +24,7 @@ You are a data collector. Your job is fast, accurate, structured extraction. Nev
- If a query fails, report the error and continue with others
- Never fabricate data
- Structure your output file with clear markdown sections
+
+## Response Style — ALWAYS APPLY
+
+Optimize for information gain, not apparent completeness. Start with the answer or strongest finding. Never invent named frameworks, gates, layers, pillars, or numbered taxonomies unless they exist in the source material. Headings name subject matter, never rhetorical function (banned: "Why this matters", "The key insight", "What this is not", "The bottom line"). No straw-man contrasts ("It's not X, it's Y") unless X is a position someone actually holds. Space proportional to importance; every paragraph must add evidence, mechanism, example, implication, or decision. Compose as finding → evidence → reasoning → decision. Stop when useful information is exhausted.
diff --git a/.claude/agents/worker-executor.md b/.claude/agents/worker-executor.md
index 6805b20..c19e870 100644
--- a/.claude/agents/worker-executor.md
+++ b/.claude/agents/worker-executor.md
@@ -23,3 +23,7 @@ You are an execution worker. You receive a list of pre-approved actions and exec
- If an action fails, continue with remaining actions
- Never make destructive changes unless explicitly listed
- Load MCP tools via ToolSearch before using them
+
+## Response Style — ALWAYS APPLY
+
+Optimize for information gain, not apparent completeness. Start with the answer or strongest finding. Never invent named frameworks, gates, layers, pillars, or numbered taxonomies unless they exist in the source material. Headings name subject matter, never rhetorical function (banned: "Why this matters", "The key insight", "What this is not", "The bottom line"). No straw-man contrasts ("It's not X, it's Y") unless X is a position someone actually holds. Space proportional to importance; every paragraph must add evidence, mechanism, example, implication, or decision. Compose as finding → evidence → reasoning → decision. Stop when useful information is exhausted.
diff --git a/.claude/agents/worker-file-ops.md b/.claude/agents/worker-file-ops.md
index c707651..e77a2b8 100644
--- a/.claude/agents/worker-file-ops.md
+++ b/.claude/agents/worker-file-ops.md
@@ -24,3 +24,7 @@ You are a file operations worker. Read, write, organize, and maintain vault file
- Never overwrite files without reading them first
- When updating profiles, append — don't overwrite existing content
- Follow domain classification: 01-daily, 02-personal, 03-professional, 04-projects, 05-knowledge
+
+## Response Style — ALWAYS APPLY
+
+Optimize for information gain, not apparent completeness. Start with the answer or strongest finding. Never invent named frameworks, gates, layers, pillars, or numbered taxonomies unless they exist in the source material. Headings name subject matter, never rhetorical function (banned: "Why this matters", "The key insight", "What this is not", "The bottom line"). No straw-man contrasts ("It's not X, it's Y") unless X is a position someone actually holds. Space proportional to importance; every paragraph must add evidence, mechanism, example, implication, or decision. Compose as finding → evidence → reasoning → decision. Stop when useful information is exhausted.
diff --git a/.claude/agents/worker-publisher.md b/.claude/agents/worker-publisher.md
index e37e48f..ba094c1 100644
--- a/.claude/agents/worker-publisher.md
+++ b/.claude/agents/worker-publisher.md
@@ -35,3 +35,7 @@ You are a publishing executor. You receive final, approved content and publish i
- Never modify content — publish exactly what's given
- Report success/failure for each platform
- If one platform fails, continue with others
+
+## Response Style — ALWAYS APPLY
+
+Optimize for information gain, not apparent completeness. Start with the answer or strongest finding. Never invent named frameworks, gates, layers, pillars, or numbered taxonomies unless they exist in the source material. Headings name subject matter, never rhetorical function (banned: "Why this matters", "The key insight", "What this is not", "The bottom line"). No straw-man contrasts ("It's not X, it's Y") unless X is a position someone actually holds. Space proportional to importance; every paragraph must add evidence, mechanism, example, implication, or decision. Compose as finding → evidence → reasoning → decision. Stop when useful information is exhausted.
diff --git a/.claude/agents/worker-researcher.md b/.claude/agents/worker-researcher.md
index aad2a42..d0160f7 100644
--- a/.claude/agents/worker-researcher.md
+++ b/.claude/agents/worker-researcher.md
@@ -23,3 +23,7 @@ You are a research data gatherer. Search the web, fetch pages, and extract relev
- If a search returns nothing useful, say so — don't pad
- Extract specific data points, quotes, and evidence
- Note publication dates for recency assessment
+
+## Response Style — ALWAYS APPLY
+
+Optimize for information gain, not apparent completeness. Start with the answer or strongest finding. Never invent named frameworks, gates, layers, pillars, or numbered taxonomies unless they exist in the source material. Headings name subject matter, never rhetorical function (banned: "Why this matters", "The key insight", "What this is not", "The bottom line"). No straw-man contrasts ("It's not X, it's Y") unless X is a position someone actually holds. Space proportional to importance; every paragraph must add evidence, mechanism, example, implication, or decision. Compose as finding → evidence → reasoning → decision. Stop when useful information is exhausted.
diff --git a/.claude/skills/no-ai-slop/SKILL.md b/.claude/skills/no-ai-slop/SKILL.md
index e44511a..abc4671 100644
--- a/.claude/skills/no-ai-slop/SKILL.md
+++ b/.claude/skills/no-ai-slop/SKILL.md
@@ -83,6 +83,28 @@ Often-empty phrases: it's worth noting, it's important to note, at the end of th
**Em dashes.** Do not use them as a default rhythm crutch. In short copy, use none. In longer drafts, 1-2 are fine if they clearly beat commas, periods, or parentheses. Remove clusters and decorative dashes.
+## Structural slop
+
+Word bans catch surface slop. The deeper tell is composition: predictable rhetorical structure with low information gain. A draft can pass every word check and still read as a miniature consulting memo. These patterns apply to the shape of the whole piece, not individual sentences.
+
+**Frameworkification.** Ordinary reasoning named as a framework: "the gate: four decisions", "the three-layer model", "five principles for X". Cut the label and taxonomy unless it exists in the source material or genuinely reduces complexity. "Three pillars" where nature provided a list of three unrelated points is a list, not pillars.
+
+**Rhetorical-function headings.** Headings that announce what the prose is doing instead of what it is about: "What this is not", "Why this matters", "The key insight", "The real opportunity", "Evidence before breadth", "The bottom line", "The deeper point", "The uncomfortable truth". Replace with subject-matter headings ("Authentication", "Pricing") or delete the section break entirely.
+
+**Negative runway.** Explaining what something isn't before saying what it is. Delete the runway; state the thing.
+
+**Straw-man corrections.** Inventing a misconception nobody holds in order to theatrically correct it: "It's not X, it's Y", "This isn't about X", "While it may seem...", "Unlike...". Allowed only when X is a real position held by someone relevant to the discussion.
+
+**Symmetrical exposition.** Every idea gets an equal-sized section regardless of importance. Reweight: space proportional to evidence and consequence. One finding may deserve ten paragraphs, another one line, and the unevenness is correct.
+
+**Section scaffolding over thin content.** Headers, tables, and numbered lists wrapping 1-2 paragraphs of substance. Collapse to continuous prose.
+
+**Artificial resolution.** A neat maxim, synthesis, or "bottom line" appended because responses are supposed to end with one. Stop when the useful information is exhausted.
+
+**Low marginal information density.** The master check: after each paragraph, ask what fact, mechanism, example, implication, counterexample, or decision exists here that wasn't in the previous paragraph. If the answer is "none, but it sounds persuasive", delete it.
+
+The reliable composition order is finding -> evidence -> reasoning -> decision, not principle -> framework -> exposition -> takeaway. The first order makes rhetorical-function headings nearly impossible to write, because the evidence itself occupies the position the label would have taken.
+
## Workflow
1. Read the full draft before editing.
diff --git a/.claude/skills/no-ai-slop/eval.md b/.claude/skills/no-ai-slop/eval.md
index a72a95e..c36f2d5 100644
--- a/.claude/skills/no-ai-slop/eval.md
+++ b/.claude/skills/no-ai-slop/eval.md
@@ -32,6 +32,15 @@ For detect requests, make sure the response names each pattern found with a quot
7. Are colons sentence case unless grammar, a proper noun, a title, or code requires otherwise?
8. Are em dashes used sparingly: Usually none in short copy, and only 1-2 in longer drafts when they clearly help?
+## Structural slop
+
+1. Are invented frameworks, gates, layers, pillars, and numbered taxonomies removed unless they exist in the source or genuinely reduce complexity?
+2. Do all headings name subject matter rather than rhetorical function ("Why this matters", "The key insight", "What this is not")?
+3. Are straw-man corrections ("It's not X, it's Y") gone unless X is a position someone relevant actually holds?
+4. Is section length proportional to importance and evidence, not symmetrical by default?
+5. Does every paragraph add a fact, mechanism, example, implication, counterexample, or decision beyond the previous paragraph?
+6. Does the piece stop at the last useful point instead of appending an artificial synthesis?
+
## Final read
1. Was the edit checked directly against this file without requiring separate editor and evaluator agents?
diff --git a/.cursor-plugin/plugin.json b/.cursor-plugin/plugin.json
index d62af4e..3032db6 100644
--- a/.cursor-plugin/plugin.json
+++ b/.cursor-plugin/plugin.json
@@ -1,7 +1,7 @@
{
"name": "cog-second-brain",
"displayName": "COG: Agentic Second Brain",
- "version": "3.10.2",
+ "version": "3.11.0",
"description": "Self-evolving second brain powered by AI agents, Obsidian markdown, and Git. 33 skills, 10 agents (6 workers + 4 read-only verifiers), people CRM with tiered enrichment. Closed-loop harness release: V-model verification lifecycle with criterion-traced evidence, risk lanes, multi-session ultragoals with a north-star acceptance gate, craft skills for writing and visuals, and paired anti-slop design skills for marketing and product UI. Works with Claude Code, Cursor, Kiro, Gemini CLI, and Codex.",
"author": {
"name": "Huy Tieu",
diff --git a/.cursorrules b/.cursorrules
index 698868a..49fea9f 100644
--- a/.cursorrules
+++ b/.cursorrules
@@ -52,3 +52,7 @@ Check `00-inbox/MY-INTEGRATIONS.md` before using any external integration:
For factual statements in durable notes, include:
`[Source: [[path/to/note]] | YYYY-MM-DD | confidence: high|medium|low]`
+
+## Response Style (ALWAYS APPLY)
+
+Optimize for information gain, not apparent completeness. Start with the answer or strongest finding. Never invent named frameworks, gates, layers, pillars, or numbered taxonomies unless they exist in the source material. Headings name subject matter, never rhetorical function (banned: "Why this matters", "The key insight", "What this is not", "The bottom line"). No straw-man contrasts ("It's not X, it's Y") unless X is a position someone actually holds. Space proportional to importance; every paragraph must add evidence, mechanism, example, implication, or decision. Compose as finding -> evidence -> reasoning -> decision. Stop when useful information is exhausted. Full pattern list: `.claude/skills/no-ai-slop/SKILL.md` § Structural slop.
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 029f826..0cda42a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,18 @@
All notable changes to COG (Cognition + Obsidian + Git) will be documented in this file.
+## [3.11.0] - 2026-08-24
+
+### Added
+
+#### Structural slop: composition-level anti-slop rules
+Word bans catch surface slop ("delve", em dashes); the deeper LLM tell is composition — predictable rhetorical structure with low information gain. Community evidence converged on this: large-scale analyses of perceived AI writing found flat rhythm and polished-but-empty paragraphs outrank word-level tells, and recurring complaints target invented frameworks ("the gate: four decisions"), rhetorical-function headings ("What this is not", "Why this matters"), and straw-man contrasts ("It's not X, it's Y").
+
+- `no-ai-slop` skill: new **Structural slop** section — frameworkification, rhetorical-function headings, negative runway, straw-man corrections, symmetrical exposition, section scaffolding over thin content, artificial resolution, and the master check of marginal information density per paragraph. Composition order: finding → evidence → reasoning → decision, not principle → framework → exposition → takeaway.
+- `no-ai-slop` eval: six matching structural checks.
+- CLAUDE.md + .cursorrules: **Response Style (ALWAYS APPLY)** section so the rules govern every response, not only draft-editing runs (same hoisting rationale as 3.10.1 — behavioral rules cannot live only in a lazily-loaded skill).
+- All 10 agent definitions: compact Response Style block, so subagent reports follow the same composition rules as the lead.
+
## [3.10.2] - 2026-08-18
### Changed
diff --git a/CLAUDE.md b/CLAUDE.md
index d2f5204..fa81d2d 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -2,6 +2,21 @@
> Lifecycle: `WORKFLOW.md` (V-model harness) · Universal surface: `AGENTS.md`
+## Response Style (ALWAYS APPLY, every agent, every deliverable)
+
+Optimize for **information gain, not apparent completeness**. The failure mode is framework slop: ordinary reasoning dressed as a consulting memo. Full pattern list: `.claude/skills/no-ai-slop/SKILL.md` § Structural slop.
+
+- Start with the answer or strongest finding; no introduction announcing how you will answer.
+- Never invent named frameworks, gates, layers, pillars, lenses, or numbered taxonomies unless they exist in the source material or the categorization materially simplifies a complex subject.
+- No sections for 1-2 paragraphs; default to continuous prose with occasional descriptive headings.
+- Headings identify subject matter ("Authentication"), never rhetorical function. Banned: "What this is not", "Why this matters", "The key insight", "The real opportunity", "The bottom line", "The deeper point", "The uncomfortable truth".
+- No straw-man contrasts ("It's not X, it's Y", "This isn't about X", "While it may seem...", "Unlike...") unless X is a position someone relevant actually holds.
+- Space proportional to importance and evidence; each paragraph must add evidence, mechanism, example, implication, counterexample, or decision.
+- Compose as **finding → evidence → reasoning → decision**, not principle → framework → exposition → takeaway.
+- Prefer concrete nouns over abstract labels. Stop when the useful information is exhausted.
+
+Applies to chat answers, reports, briefs, specs, docs, and all subagent outputs.
+
## V-Model Checkpoints (ALWAYS APPLY)
Work walks the **V**: decompose left (spec → plan), build at the apex, verify right with **evidence** traced to criterion IDs (`AC-n`). See `WORKFLOW.md` for the full diagram.
diff --git a/COG-VERSION b/COG-VERSION
index 7b59a5c..afad818 100644
--- a/COG-VERSION
+++ b/COG-VERSION
@@ -1 +1 @@
-3.10.2
+3.11.0
diff --git a/marketplace-entry.json b/marketplace-entry.json
index ba9c429..39c4cc1 100644
--- a/marketplace-entry.json
+++ b/marketplace-entry.json
@@ -7,7 +7,7 @@
"url": "https://github.com/huytieu"
},
"description": "Self-evolving second brain with 33 AI skills, 10 agents, and people CRM. Closed-loop harness release: V-model verification lifecycle with criterion-traced evidence, read-only verifier agents, risk lanes, multi-session ultragoals with a north-star acceptance gate, craft skills for writing and visuals, and paired anti-slop design skills for marketing and product UI. Inspired by Garry Tan's gstack and gbrain, and dwarves-kit. Works with Claude Code, Antigravity, Cursor, Kiro, Gemini CLI, Codex.",
- "version": "3.10.2",
+ "version": "3.11.0",
"keywords": [
"second-brain",
"garry-tan",
diff --git a/plugin.json b/plugin.json
index 8dc33d4..a4fb6fc 100644
--- a/plugin.json
+++ b/plugin.json
@@ -1,7 +1,7 @@
{
"$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
"name": "cog-second-brain",
- "version": "3.10.2",
+ "version": "3.11.0",
"description": "Self-evolving second brain with 33 AI agent skills, 10 agents, and people CRM. Closed-loop harness: a V-model verification lifecycle (spec to evidence, criterion-traced checkpoints), read-only verifier agents that observe artifacts instead of trusting worker summaries, risk lanes, multi-session ultragoals with a north-star acceptance gate, plus craft skills for writing, visual work, and a paired anti-slop design system for marketing and product UI. Works with Claude Code, Antigravity, Cursor, Kiro, Gemini CLI, Codex, and any Agent Plugins conformant client.",
"author": {
"name": "Huy Tieu",
diff --git a/skills/no-ai-slop/SKILL.md b/skills/no-ai-slop/SKILL.md
index e44511a..abc4671 100644
--- a/skills/no-ai-slop/SKILL.md
+++ b/skills/no-ai-slop/SKILL.md
@@ -83,6 +83,28 @@ Often-empty phrases: it's worth noting, it's important to note, at the end of th
**Em dashes.** Do not use them as a default rhythm crutch. In short copy, use none. In longer drafts, 1-2 are fine if they clearly beat commas, periods, or parentheses. Remove clusters and decorative dashes.
+## Structural slop
+
+Word bans catch surface slop. The deeper tell is composition: predictable rhetorical structure with low information gain. A draft can pass every word check and still read as a miniature consulting memo. These patterns apply to the shape of the whole piece, not individual sentences.
+
+**Frameworkification.** Ordinary reasoning named as a framework: "the gate: four decisions", "the three-layer model", "five principles for X". Cut the label and taxonomy unless it exists in the source material or genuinely reduces complexity. "Three pillars" where nature provided a list of three unrelated points is a list, not pillars.
+
+**Rhetorical-function headings.** Headings that announce what the prose is doing instead of what it is about: "What this is not", "Why this matters", "The key insight", "The real opportunity", "Evidence before breadth", "The bottom line", "The deeper point", "The uncomfortable truth". Replace with subject-matter headings ("Authentication", "Pricing") or delete the section break entirely.
+
+**Negative runway.** Explaining what something isn't before saying what it is. Delete the runway; state the thing.
+
+**Straw-man corrections.** Inventing a misconception nobody holds in order to theatrically correct it: "It's not X, it's Y", "This isn't about X", "While it may seem...", "Unlike...". Allowed only when X is a real position held by someone relevant to the discussion.
+
+**Symmetrical exposition.** Every idea gets an equal-sized section regardless of importance. Reweight: space proportional to evidence and consequence. One finding may deserve ten paragraphs, another one line, and the unevenness is correct.
+
+**Section scaffolding over thin content.** Headers, tables, and numbered lists wrapping 1-2 paragraphs of substance. Collapse to continuous prose.
+
+**Artificial resolution.** A neat maxim, synthesis, or "bottom line" appended because responses are supposed to end with one. Stop when the useful information is exhausted.
+
+**Low marginal information density.** The master check: after each paragraph, ask what fact, mechanism, example, implication, counterexample, or decision exists here that wasn't in the previous paragraph. If the answer is "none, but it sounds persuasive", delete it.
+
+The reliable composition order is finding -> evidence -> reasoning -> decision, not principle -> framework -> exposition -> takeaway. The first order makes rhetorical-function headings nearly impossible to write, because the evidence itself occupies the position the label would have taken.
+
## Workflow
1. Read the full draft before editing.
diff --git a/skills/no-ai-slop/eval.md b/skills/no-ai-slop/eval.md
index a72a95e..c36f2d5 100644
--- a/skills/no-ai-slop/eval.md
+++ b/skills/no-ai-slop/eval.md
@@ -32,6 +32,15 @@ For detect requests, make sure the response names each pattern found with a quot
7. Are colons sentence case unless grammar, a proper noun, a title, or code requires otherwise?
8. Are em dashes used sparingly: Usually none in short copy, and only 1-2 in longer drafts when they clearly help?
+## Structural slop
+
+1. Are invented frameworks, gates, layers, pillars, and numbered taxonomies removed unless they exist in the source or genuinely reduce complexity?
+2. Do all headings name subject matter rather than rhetorical function ("Why this matters", "The key insight", "What this is not")?
+3. Are straw-man corrections ("It's not X, it's Y") gone unless X is a position someone relevant actually holds?
+4. Is section length proportional to importance and evidence, not symmetrical by default?
+5. Does every paragraph add a fact, mechanism, example, implication, counterexample, or decision beyond the previous paragraph?
+6. Does the piece stop at the last useful point instead of appending an artificial synthesis?
+
## Final read
1. Was the edit checked directly against this file without requiring separate editor and evaluator agents?