fill
Skill25 starsChanged 8 months agoMITView on GitHub
Guide content filling for scaffolded products. Walks sections, asks domain questions, writes expertise into files. Use after /create in 'scaffold' state, or when the creator says 'fill', 'add content', or 'help me complete this'.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- askoulwassim/perfect-placementtheir own version, first seen Oct 15, 2026
- leanEthereum/leanSpectheir own version, first seen Oct 8, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- leanEthereum/leanSpec★ 143same name
- askoulwassim/perfect-placement★ 3same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
Show its rules and text
We split this skill into the rules it gives an agent
35 rules. About 0 are common, 17 are distinctive, and 18 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈17
Distinctive · estimate
This author's own choices. Worth a look.
18
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 21
When to use: After /create has generated a scaffold.
- Distinctive: line 23
When NOT to use: If the product doesn't exist yet (use /create first).
- Very specific: line 34
If $ARGUMENTS provided, use as product slug → look in workspace/{slug}/
- Very specific: line 35
If workspace/{slug}/ does not exist → "Product {slug} not found in workspace/.
- Very specific: line 36
If not, glob workspace/*/ and list products in scaffold/content state
- Very specific: line 37
If no products in scaffold/content state found → "No products ready to fill.
- Very specific: line 38
If multiple products found, ask which one to fill 1b.
- Very specific: line 43
Read .meta.yaml from product directory → get type, state, mcs_target, AND the intent_declaration block if present.
- Very specific: line 44
Intent-aware calibration: If intent_declaration is present, extract engine_parsed.{depth, nature, delivery_mechanism}.
- Very specific: line 45
Legacy fallback: If intent_declaration is absent (legacy product) OR if intent_declaration.mode == legacy_fallback with all engine_parsed …
- Very specific: line 46
Maintain creator persona: Read creator.yaml → adapt language, depth, and examples to profile.type and technical_level throughout this …
- Very specific: line 76
Load scout report: if .meta.yaml has intent_declaration.scout_source field (canonical location) OR .meta.yaml root-level scout_source field …
- Very specific: line 78
Read the most recent sibling's .meta.yaml → intelligence and state.overall_score
- Distinctive: line 79
If sibling substance_score < 50: "Your last product in {domain} scored {substance}% substance.
- Distinctive: line 80
If sibling substance_score >= 70: "Your {domain} products have strong substance ({substance}%).
- Very specific: line 83
Intelligence gap check: If .meta.yaml has NO scout_source AND product targets MCS-2+:
- Distinctive: line 87
If creator continues without scout, skip research injection steps in section walker
- Very specific: line 88
Structured input for choices: If creator.yaml → preferences.workflow_style = guided, use AskUserQuestion for all choice points …
- Very specific: line 89
Read fill-protocol.md: Read ${CLAUDE_SKILL_DIR}/references/fill-protocol.md now — it contains the full execution protocol for all phases …
- Distinctive: line 94
Follow the content pack's walker entry point (§1) which handles:
- Distinctive: line 103
After the SELF-CLONE walker completes, skip to Completion (step 13 standard flow is bypassed)
- Very specific: line 111
Ask archetype if not in .meta.yaml: Sequential Pipeline / Parallel Fan-Out / Conditional Router / Iterative Refinement / Hierarchical …
- Very specific: line 112
Role-aware composition (entity-ontology.md §AGENT_ROLES): Instead of generic "how many specialists?", present the 7 functional roles: "Each …
- Distinctive: line 114
Ask: "Name each specialist, their domain, and their role.
- Distinctive: line 116
Ask entity lifecycles: "What objects flow through this squad?
- Very specific: line 129
If agent has a mind-clone source in minds/: load minds/{name}/cognitive-model.md and use as knowledge substrate
- Distinctive: line 135
Ask per agent: "What are the 3-5 most important things {agent_name} does?
- Distinctive: line 139
Ask: "Walk me through the main workflow from start to finish.
- Distinctive: line 157
If any component is missing, flag it: "Your squad is missing {component}.
- Very specific: line 158
Completion: After all phases, update .meta.yaml with fill_metrics and transition state to content.
- Distinctive: line 163
Use the selected role to determine:
- Distinctive: line 175
If evidence suggests LLM judgment needed → surface heuristic from §HEURISTICS as coaching
- Distinctive: line 202
Before promoting to "content" state:
- Distinctive: line 208
Auto-validate honesty rule: When the silent MCS-1 check at completion finds failures, the creator-facing message MUST name them.
- Distinctive: line 227
When context is compressed, preserve:
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Content Filler
Extract domain expertise from creator conversation and inject into product files.
When to use: After /create has generated a scaffold. Product state should be "scaffold" or "content".
When NOT to use: If the product doesn't exist yet (use /create first). If the product is already validated (edits will regress state).
Full protocol details (section walker, sparring, extraction modes, type-specific coaching, research injection, progress tracking, completion flow) are in:
${CLAUDE_SKILL_DIR}/references/fill-protocol.md— Read this file before executing the section walk.
Activation Protocol
- Shared preamble: Load
references/quality/activation-preamble.md— context assembly, persona adaptation, deterministic routing rules. - Identify target product:
- If
$ARGUMENTSprovided, use as product slug → look inworkspace/{slug}/ - If
workspace/{slug}/does not exist → "Product{slug}not found in workspace/. Run/createfirst or checkworkspace/for available slugs." - If not, glob
workspace/*/and list products in scaffold/content state - If no products in scaffold/content state found → "No products ready to fill. Run
/create {type}to scaffold one first." - If multiple products found, ask which one to fill
1b. Mode selection (Express vs Guided). Read
creator.yaml → preferences.workflow_style. Resolve the flow mode: --expressflag ORworkflow_style == "autonomous"→ Express mode. Skip the three Discovery front-loaded questions, skip the Pitfall Check interactive prompt, skip Extraction Mode menu (default to Standard), skip the Deepening menu after each section, run the section walker with type defaults, and fall straight through to Completion. Interactive prompts are replaced by type-based defaults + one-line advisory notes. Proactive brainstorm prompts are suppressed.workflow_style == "guided"or missing → Guided mode (default). Run the full protocol as documented in fill-protocol.md. Recordfill_config.mode: express | guidedin.meta.yamlso later skills can reason about the creator's chosen rhythm.
- If
- Read
.meta.yamlfrom product directory → get type, state, mcs_target, AND theintent_declarationblock if present.- Intent-aware calibration: If
intent_declarationis present, extractengine_parsed.{depth, nature, delivery_mechanism}. These three fields drive section walker routing, tone calibration, and question shape — seefill-protocol.md → INTENT-AWARE CALIBRATIONfor the full rubric. Recordfill_config.intent_aware: truein.meta.yamlwhen the calibration fires. - Legacy fallback: If
intent_declarationis absent (legacy product) OR ifintent_declaration.mode == legacy_fallbackwith all engine_parsed fields null, fall back to type-based defaults. Emit one advisory line: "This product lacks intent metadata — /fill will use type-based defaults. Re-run /create to unlock intent-aware filling." Recordfill_config.intent_aware: false.
- Intent-aware calibration: If
- Maintain creator persona: Read
creator.yaml→ adapt language, depth, and examples toprofile.typeandtechnical_levelthroughout this skill's execution. A developer gets code examples; a domain expert gets plain language. - Load UX stack (in order):
references/quality/engine-voice-core.md— the micro voice contract carried through every question, section signal, and sparring line in /fill. This is where the Creator spends the most time; the voice cannot drift.references/ux-experience-system.md§1 Context Assembly (build creator context), §2.2 Archetype-Aware Insights (adapt emphasis to creator goals), §2.3 Moment Awareness (mid-fill coaching)references/ux-vocabulary.md— translate terms in any creator-facing output Vocabulary enforcement (mandatory): Every question, section signal, sparring challenge, and progress message passes through ux-vocabulary.md translation before reaching the creator. Internal terms (Sparring, Pitfall, MCS, DNA tier, extraction mode) are replaced for non-dev creators. For dev/hybrid creators, terms may appear but always with context ("MCS-2 quality checks — these verify craft and expertise depth").references/quality/engine-voice.md— full voice substrate. Load when composing section quality signals, sparring pressure, milestone celebrations, or brand moments.references/quality/exemplar-outputs.mdsections E4 and E5 only — the section question and sparring exemplars. These show the exact visual rhythm and warmth your questions must carry: progress sandwich (bar → celebration → question → escape hatch), warm framing, 💡 scaffolding offers.references/ux-experience-system.md§11 Deep Elicitation Protocol — MANDATORY for all AskUserQuestion calls. Apply: experience-based questions (not abstractions), one question at a time, mirror to confirm, escalate progressively, offer scaffolding when stuck, celebrate depth. /fill is where the creator spends the MOST time. The experience must be warm coaching, not interrogation. Adapt: beginners get encouragement + examples. Experts get peer-level sparring. Celebrate section completions with progress visibility ("4/7 sections filled. Core identity locked in."). Hyper-personalize using creator.yaml fields — name, goals, expertise areas. 4b. Load proactives: Loadreferences/engine-proactive.md— wire #1 (pipeline guidance: after fill completes, guide to /validate), #17 (lost creator: if stuck on a section for 3+ questions, offer /think or skip), #19 (error recovery: if fill encounters malformed scaffold, suggest re-scaffolding).
- Load product spec from
references/product-specs/{type}-spec.md - Load product DNA from
product-dna/{type}.yaml6b. Load entity ontology (squad/system/agent/workflow/minds): If type ∈ {squad, system, agent, minds, workflow}, readreferences/entity-ontology.md. This substrate drives semantic elicitation:- §AGENT_ROLES: suggest specific roles during composition discovery, adapt questions per role (EXECUTOR gets tool/output questions, ADVISOR gets reasoning questions, VALIDATOR gets criteria questions)
- §SQUAD_ANATOMY: verify all 8 components are addressed during fill — flag any missing component before completion
- §COMPOSITION: know what this type can compose with and what's forbidden — guide toward valid compositions
- §HERITAGE: explain to the creator why certain DNA patterns apply (squad inherits from agent lineage)
- §INTELLIGENCE_GRADIENT: calibrate autonomy level in routing questions — workflows get deterministic, squads get judgment-based
- §WORKFLOW_VS_SQUAD: when filling a workflow, enforce YAML-decides (not LLM routing). When filling a squad, surface LLM judgment points
- §HEURISTICS: if evidence suggests wrong type (e.g., squad with 1 useful agent), surface demotion heuristic as coaching
- §RUNTIME_BEHAVIOR: calibrate token budget expectations and explain context isolation to the creator
- §ISOMORPHIC: use human cognitive analogies to explain product role to non-dev creators
- For type=system ONLY: §SYSTEM_ENGINES — the 13 functional gears. Walk each active gear during fill, ask for concrete implementation decisions, verify counterpart coupling
- §INTELLIGENCE_PIPELINE — governs how scout research ACTIVELY enters product content (not passively). For EVERY section: check if scout report has relevant findings → propose content from research → creator validates/refines → sparring challenges generic answers. This is the soul of the Engine — condensing intelligence, not just filling templates.
- §FOUNDATIONAL_THESIS — for EVERY product, ask: "What should this NEVER be able to do?" The answer generates the intelligence flow. Use to guide frontmatter restrictions (denied-tools, paths: scoping).
- §TRANSVERSAL_AXES — calibrate Nature (procedural/advisory/executive/orchestrative) and Depth (surface/functional/cognitive) during fill. A skill CAN be cognitive-depth. Don't assume type=depth.
- §CONSTITUTIONAL_PRIMITIVES — for agents and minds, suggest criticalSystemReminder when the product must maintain identity over long conversations (50+ turns). Especially for cognitive minds.
- §COMPOSITION_PRINCIPLES — when filling a squad or system, apply convergence-by-independence (specialists don't see each other's work until synthesis) and symbiosis rules.
- Load
domain-map.mdif it exists (from /map) → use as knowledge source - Load scout report if
.meta.yamlhasintent_declaration.scout_sourcefield (canonical location) OR.meta.yamlroot-levelscout_sourcefield (legacy fallback) → readworkspace/{scout_source}for research context. Extract: baseline (Section 1), gaps (Section 2), research findings (Section 4). This intelligence drives research injection in the section walk. 8b. Domain intelligence back-reference: ReadSTATE.yaml → workspace.products[]. Find products in the sameintelligence.domainas the current product. If any exist withintelligence.value_scorepopulated:- Read the most recent sibling's
.meta.yaml → intelligenceandstate.overall_score - If sibling substance_score < 50: "Your last product in {domain} scored {substance}% substance. Focus on depth and real examples this time."
- If sibling substance_score >= 70: "Your {domain} products have strong substance ({substance}%). Maintain this depth."
- This closes the feed-back loop: /validate's output informs /fill's approach.
- Read the most recent sibling's
- Scan product files for template placeholders and WHY comments
- Intelligence gap check: If
.meta.yamlhas NOscout_sourceAND product targets MCS-2+:- Show: "No scout report found. Without baseline intelligence, /fill works in creator-knowledge-only mode — no research proposals, no baseline comparison."
- Suggest: "Run
/scout {inferred_domain}first for research-backed filling. Or continue with your expertise." - Record
fill_config.scout_available: falsein .meta.yaml - If creator continues without scout, skip research injection steps in section walker
- Structured input for choices: If
creator.yaml → preferences.workflow_style = guided, useAskUserQuestionfor all choice points (enhance/replace/skip, deepening method selection, continue/save progress). Use plain text only for open-ended domain knowledge extraction. - Read fill-protocol.md: Read
${CLAUDE_SKILL_DIR}/references/fill-protocol.mdnow — it contains the full execution protocol for all phases below. 12b. SELF-CLONE branch (cognitive minds with sub_type=self only). If.meta.yaml.type == "minds" AND .meta.yaml.minds_sub_type == "self":- Load the SELF-CLONE content pack: Read
references/fill-content-packs/self-clone.md - The content pack replaces the standard Discovery, Extraction Mode, and generic Section Walk phases
- Follow the content pack's walker entry point (§1) which handles:
- Mode selection (distillation vs elicitation based on corpus density)
- Distillation pass (if applicable) with creator confirm/refine/reject per entry
- Gap elicitation with the ordered question sequences per dimension
- Dimensional routing (primary + secondary tagging)
- Typology mirror (post-elicitation, three-step protocol)
- Coherence diff generation
- Honesty floor gates (counter-proofs, signed incaptable list, uncapturable decisions)
- Writing populated content into the cognitive mind layer files
- After the SELF-CLONE walker completes, skip to Completion (step 13 standard flow is bypassed)
- The Sparring, Checkpointing, and Progress phases from the standard flow still apply during the walker — checkpoint after every two dimensions
- Record
fill_config.self_clone: truein.meta.yaml - Do not run this branch for non-SELF cognitive minds or non-minds products — all other types continue to step 12c or 13
12c. SQUAD ORGANISM branch (type=squad only).
If
.meta.yaml.product.type == "squad": - This branch replaces the standard section walker. Squads are organism-level products — the standard single-file walk is insufficient. The squad walker iterates across the full directory tree.
- Phase 1 — Composition Discovery (if not already answered in /create):
- Ask archetype if not in
.meta.yaml: Sequential Pipeline / Parallel Fan-Out / Conditional Router / Iterative Refinement / Hierarchical Delegation - Role-aware composition (entity-ontology.md §AGENT_ROLES): Instead of generic "how many specialists?", present the 7 functional roles: "Each specialist has a functional ROLE that determines tools and output. Select from: EXECUTOR (acts on files), SPECIALIST (deep analysis, read-only), ORCHESTRATOR (coordinates agents), ROUTER (classifies + directs), ADVISOR (thinks, never acts), VALIDATOR (checks quality), TRANSFORMER (converts formats)." Ask: "Name each specialist, their domain, and their role. Example: 'researcher=SPECIALIST, strategist=ADVISOR, executor=EXECUTOR, reviewer=VALIDATOR'." Record role assignment per agent for Phase 3 frontmatter generation.
- Ask entity lifecycles: "What objects flow through this squad? (e.g., a CAMPAIGN, a LEAD, an EXPERIMENT). Describe the lifecycle stages."
- Ask archetype if not in
- Phase 2 — Orchestrator Layer (kernel/):
- Walk
kernel/orchestration.md— fill orchestration protocol. Ask: "How does the coordinator decide what to do? Describe its decision cycle step by step." - Walk
kernel/elicitation-engine.yaml— fill progressive elicitation rules. Ask: "What questions does the squad ask the user before acting? How does it adapt based on answers?" - Walk
kernel/intelligence-matrix.yaml— fill scoring/routing intelligence. Ask: "How does the squad score which agents are most relevant for a given request?"
- Walk
- Phase 3 — Specialist Agents (agents/):
- For EACH agent file in
agents/*.md:- Walk identity, role, tools, escalation rules
- Map to AGENT ROLE from entity-ontology.md §AGENT_ROLES. The role assigned in Phase 1 determines:
- Tool boundary: EXECUTOR=Write/Edit/Bash, SPECIALIST/VALIDATOR=Read-only, ORCHESTRATOR=Agent-only (no Write/Edit), ADVISOR=denied-tools
- Handoff format: EXECUTOR→artifacts, ADVISOR→judgment, ORCHESTRATOR→routing decisions, VALIDATOR→score+verdict, TRANSFORMER→converted output
- Frontmatter: set
allowed-toolsordenied-toolsper role mapping in the agent .md file - Question shape: EXECUTOR→"What does it produce?", ADVISOR→"What reasoning framework?", VALIDATOR→"What criteria?", TRANSFORMER→"What's the input/output format?"
- If agent has a mind-clone source in
minds/: loadminds/{name}/cognitive-model.mdand use as knowledge substrate - Apply D1 (activation protocol), D2 (anti-patterns ≥5), D14 (graceful degradation) per agent
- Sparring: "What would this specialist REFUSE to do? What's outside its boundary?"
- For EACH agent file in
- Phase 4 — Task Registry (tasks/):
- Walk
tasks/task-registry.yaml— for each specialist, define 3-10 granular tasks with:- task_id, description, assigned_agent, input_schema, output_schema, quality_gate
- Ask per agent: "What are the 3-5 most important things {agent_name} does? For each, what input does it need and what output does it produce?"
- Walk
- Phase 5 — Chains & Workflows (chains/, workflows/):
- Walk
chains/chain-registry.yaml— define micro-workflows (2-4 agent sequences with cadences) - Walk
workflows/*.md— define full pipelines mapping lifecycle stages to agents - Ask: "Walk me through the main workflow from start to finish. Which agent handles each step? What passes between them?"
- Walk
- Phase 6 — Routing & Handoff (config/):
- Walk
config/routing-table.md— fill declarative routing rules (IF intent → agent) - Calibrate routing per §WORKFLOW_VS_SQUAD: Squad routing involves LLM judgment — the orchestrator reads rules BUT interprets contextually. Ensure routing questions surface the judgment points: "When the input is ambiguous, how does the orchestrator decide?" This is what makes it a squad, not a workflow.
- Use §INTELLIGENCE_GRADIENT to calibrate autonomy: "How much should the orchestrator decide autonomously vs. escalate to the human? Where on the deterministic↔autonomous spectrum should routing sit?"
- Walk
config/handoff-protocol.md— define handoff envelope format (XML or structured) - Sparring: "What happens when the input doesn't match any route? What's the fallback?"
- Walk
- Phase 7 — Quality & Testing (tests/):
- Walk test scenarios: happy path, edge case, adversarial, agent failure recovery
- Anatomy completeness check (entity-ontology.md §SQUAD_ANATOMY): Before transitioning to content state, verify all 8 squad anatomy components have been addressed:
- Agent Roster — at least 2 agent files in agents/ with substantive content
- Routing Table — config/routing-table.md has routing rules (not just placeholder)
- Handoff Protocols — config/handoff-protocol.md has envelope format
- Workflows — at least 1 workflow with steps defined in workflows/
- Skills-as-Instruments — skills/ directory exists (advisory — some squads don't need shared skills)
- Checklists — SQUAD.md has quality/checklist section with ≥2 items
- Templates — kernel/ has output format standards
- Escalation — SQUAD.md has escalation section with confidence thresholds If any component is missing, flag it: "Your squad is missing {component}. Without it, {consequence}. Want to add it now or defer?"
- Completion: After all phases, update
.meta.yamlwithfill_metricsand transition state tocontent. - Record
fill_config.squad_organism: truein.meta.yaml12d. Agent ontology-aware elicitation (type=agent only, non-squad). If type == "agent" AND branch 12c did NOT fire: - Entity ontology already loaded at step 6b
- At the START of the section walk, ask the role question: "What functional role does this agent fill? EXECUTOR (acts on files), SPECIALIST (analysis only), ADVISOR (thinks, never acts), VALIDATOR (checks quality), TRANSFORMER (converts formats)?"
- Use the selected role to determine:
- Which tools to suggest in the frontmatter section (EXECUTOR→Write/Edit/Bash, SPECIALIST→Read-only, ADVISOR→denied-tools)
- What output to expect (artifacts vs reports vs reasoning)
- What questions to emphasize (EXECUTOR→"What does it produce?", ADVISOR→"What reasoning framework?", VALIDATOR→"What criteria does it check?")
- Record role in
.meta.yaml → agent_rolefor /validate consumption 12e. Workflow ontology-aware elicitation (type=workflow only). If type == "workflow": - Entity ontology already loaded at step 6b
- Enforce §WORKFLOW_VS_SQUAD: workflows use SKILLS (not agents), have FIXED sequences (not LLM routing), are DETERMINISTIC (same input → same execution path)
- During the section walk, ask:
- "What skills does each step invoke?" (NOT "what agents" — workflows don't use agents)
- "Is the sequence always the same regardless of input?" (If answer is "it depends" → suggest squad: "This sounds like it needs contextual routing — that's a squad, not a workflow. Want to switch?")
- "What are the gate conditions between steps?" (pass/fail criteria for proceeding)
- If evidence suggests LLM judgment needed → surface heuristic from §HEURISTICS as coaching
- Load the SELF-CLONE content pack: Read
- Begin section-by-section guided extraction (see fill-protocol.md → DISCOVERY PHASE)
Core Flow (routing table — details in fill-protocol.md)
| Phase | What happens | Protocol section |
| Discovery | 3 front-loaded questions (audience, differentiator, use case) | DISCOVERY PHASE |
| Acceptance Criteria | Define 3 machine-verifiable criteria before section walk | ACCEPTANCE CRITERIA CAPTURE |
| Extraction Mode | Select mode (Standard/Socratic/Adversarial/Archaeological/Council) for MCS-2+ | EXTRACTION MODE SELECTION |
| Pitfall Check | Load meta/pitfalls/pitfalls.json, surface type-relevant warnings | PITFALL CHECK |
| Type Coaching | Platform-specific coaching (claude-md, minds, hooks, skill/agent, system/bundle) | TYPE-SPECIFIC COACHING |
| Section Walk | Per-section: read WHY, check existing, inject research, ask, write, quality signal | SECTION WALKER |
| Sparring | 3 challenges per major section (Generic Test, Inversion, Proof) — mandatory MCS-2+ | SPARRING PROTOCOL |
| Deepening | 5 elicitation methods offered as optional menu after each major section | ELICITATION DEEPENING |
| Checkpointing | Write fill_progress to .meta.yaml after every section (compact-safe) | MID-FILL STATE PERSISTENCE |
| Progress | Pulse every 2-3 sections: filled/total + live MCS estimate | PROGRESS TRACKER |
| Conversational | Narrative extraction mode for non-dev creators | CONVERSATIONAL MODE |
| Size Check | Auto-split primary file if >4K chars; suggest @include for claude-md; enforce calibration budget — if intent-aware calibration set a size target, warn when exceeded | INSTRUCTION SIZE OPTIMIZATION |
| Completion | Promote to content, auto-validate MCS-1, report result honestly (pass → celebrate; fail → name the failures), clear fill_progress | AFTER ALL SECTIONS FILLED |
Quality Gate
Before promoting to "content" state:
- Primary file has at least 3 sections with substantive content
- No sections contain ONLY placeholder/template text
- README.md follows
templates/readme/README.md.templatestructure: Hero (problem hook), Install, "Is this for me?", Quick Start, Features (with depth — heading + what + why per feature), Use Cases (scenario table), How It Works, type-specific section, Requirements, Compatibility, Language, License, Footer. Trilingual (EN, PT-BR, ES) with anchor navigation. Each language block has FULL content — not just translated headers. - At least one domain-specific insight encoded (not generic)
Auto-validate honesty rule: When the silent MCS-1 check at completion finds failures, the creator-facing message MUST name them. "Looking good" is forbidden when checks fail. Use: "Content captured. {N} structural items need attention before validation passes: {list}. Run /validate --fix to resolve."
Anti-Patterns
- Robotic Q&A — Don't be a questionnaire. Be conversational. Reference previous answers. Connect dots between sections.
- Overwriting creator edits — If creator has already manually edited a section, don't overwrite. Ask: "This section has content. Enhance, replace, or skip?"
- Generic fill — If a section reads like any AI could write it, push: "What's YOUR specific approach here? What would surprise someone reading this?"
- Section overload — Don't try to fill everything in one session. After 5-6 sections, offer: "We covered the core. Want to continue or save progress and come back?"
- Ignoring domain-map — If domain-map.md exists, USE it. Don't re-ask questions already answered there.
- Thin sections without flagging — Never leave a <50 word section without telling the creator it's thin. Section quality signals are mandatory.
- Skipping acceptance_criteria — Must-have capture is NOT optional. Without acceptance criteria, /validate has nothing goal-backward to check.
- Forgetting persona mid-fill — Re-read creator.yaml persona BEFORE each major section. A developer gets different questions than a marketer, even for the same section.
Compact Instructions
When context is compressed, preserve:
- Target product slug, type, current phase (scaffold/content)
fill_progress.last_completed_sectionandfill_progress.sections_filled- Active extraction mode (
fill_config.extraction_mode) - Sparring state:
sparring.sections_challenged,sparring.skipped - Scout availability (
fill_config.scout_available) - Any blocking issues found during mid-fill quality signals
- Next section to fill (resume from
last_completed_section + 1)
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

