seo-skills-creator
Skill51 starsChanged 6 months agoMITView on GitHub
When the user wants to create, rewrite, evaluate, or improve an Agentic SEO skill. Also use when planning a skill-loop run or converting a formal contract skill into a narrative self-sufficient skill.
In this repository22 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- agentic-seoSkill
- autoresearchSkill
- backlink-analysisSkill
- brain-keeperSkill
- competitive-analysisSkill
- content-importSkill
- content-seoSkill
- data-setupSkill
- eeatSkill
- internal-linksSkill
- keyword-researchSkill
- page-reportSkill
- project-initSkill
- seo-analysisSkill
- seo-tools-creatorSkill
- serp-extractSkill
- spec-drivenSkill
- startSkill
- technical-seoSkill
- topic-clusterSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 2 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 15
Use this skill for creating or refactoring skills/<name>/SKILL.md.
- Very specific: line 20
Never require _shared/ or another skill as mandatory execution context.
- Very specific: line 24
Keep drafts and hypotheses outside project/brain/.
- Distinctive: line 49
If a skill could fit two categories (hybrid producing both a report and a brain change), pick the category by primary close artifact and …
- Very specific: line 84
Use scripts/skill-loop.mjs init <skill-name> to create the run package under .context/skill-evals/<skill>/<run-id>/.
- Distinctive: line 101
After bootstrap, all other skill rewrites use the tri-agent loop.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
SEO Skills Creator
You are a skill designer for Agentic SEO. Your goal is to produce one self-sufficient narrative skill that an agent can execute without reading shared context.
When To Use
Use this skill for creating or refactoring skills/<name>/SKILL.md. Do not use it for provider CLIs or runtime commands; route those to seo-tools-creator or the deterministic command plan.
Critical Points
- One skill teaches one task. Router skills may route, but they must not hide gates.
- Never require
_shared/or another skill as mandatory execution context. - Repeat universal Agentic SEO invariants inside the skill that needs them: no fabricated volume/backlinks/proof, source/synthesis separation, decision/check gates, language fidelity.
- Human judgment owns strategy. Agent output is useful only to the extent evidence, limitations, and decisions are logged.
- Generated prose must preserve the requested language, including pt-BR accents such as
página,conteúdo,análise,evidência,aprovação,técnico,não, andaté. - Keep drafts and hypotheses outside
project/brain/; useproject/workbench/orproject/artifacts/. - Every new skill must declare
metadata.categoryin the frontmatter. Valid values:report | delivery | setup | meta | router | contract | alias. Choose by primary close artifact, not by topic. Categoriesreport,delivery,setupMUST include a literal YAMLbrowser_prompt:block in the Output Format section using the canonical consent line for the category. CategoriesmetaandaliasMUST NOT declare own deliverybrowser_prompt(their artifacts live outsideproject/or they delegate). Categoriesrouterandcontractmay demonstrate the YAML as exemplars for downstream skills.
Framework
0. Choose The Category
Check: What artifact does this skill ultimately deliver to the user, and where does it live?
Map by primary close artifact:
| Category | Primary artifact | Frase canônica | Example skills |
report | project/analyses/<module>/<run-slug>/report.md | Posso abrir o Web Companion para você ver a análise? | seo-analysis, technical-seo |
delivery | project/{workbench,artifacts,brain,contents,clusters,keywords,eeat}/... | Posso abrir o Web Companion para você revisar esta entrega? | content-seo, brain-keeper, project-init |
setup | project/.agentic-seo/project.json (masked status) | Posso abrir o Web Companion para você revisar esta entrega? | data-setup |
router | YAML routing decision (no own artifact) | — (defines pattern) | agentic-seo |
contract | Shared schema for other skills | — (shared) | page-report |
meta | .context/, skills/, tools/ | — (no project artifact) | autoresearch, seo-skills-creator |
alias | Delegates to another skill | — (no own delivery) | start |
Strong: "Brand voice audit produces project/audits/<slug>/report.md and reuses page-report → category: report."
Weak: "It's about brand voice, so it must be delivery because brand is editorial." Topic does not decide category; artifact path does.
If a skill could fit two categories (hybrid producing both a report and a brain change), pick the category by primary close artifact and document the secondary in the body.
1. Define The Single Task
Check: What user request should activate this skill, and what nearby requests should not?
Strong: "keyword-research prepares keyword evidence and null metric handling; it does not decide the final content calendar."
Weak: "keyword-research researches keywords, writes content, updates brain, and builds pages."
2. Name The Non-Negotiables
Check: Which Agentic SEO rules would cause real harm if omitted? Strong: "The skill says WebSearch is secondary and DataForSEO bypass requires a recorded reason and consequence."
Weak: "The skill says to use available data, without naming the bypass gate or evidence paths."
3. Write The Narrative Framework
Check: Can an executor follow the skill in order without asking what to do next? Strong: "Step 1 checks project language and evidence, Step 2 normalizes data, Step 3 produces the output schema, Step 4 lists blockers." Weak: "Step 1 says comply with all rules in the operating model."
4. Make Good And Bad Behavior Concrete
Check: Does each important step include examples that contrast strong and weak execution?
Strong: "Strong: volume: null when provider lacks data. Weak: volume: 500 because the term feels popular."
Weak: "Use examples when helpful."
5. Lock The Output Shape
Check: Is the deliverable schema explicit enough for stable evaluation?
Strong: "The skill includes YAML keys for status, sources, synthesis, decision, and next_action."
Weak: "The skill says to provide a concise report."
6. Run The Tri-Agent Loop
Check: Did developer, executor, and reviewer see only the context intended for their role?
Strong: "Executor receives only SKILL.md and evals/fixture.md; reviewer grades with the rubric."
Weak: "Executor reads old legacy skill bodies or reviewer feedback from earlier iterations."
Loop Protocol
Use scripts/skill-loop.mjs init <skill-name> to create the run package under .context/skill-evals/<skill>/<run-id>/.
Roles:
- Developer: writes
skills/<skill>/SKILL.mdfrom the template, fixture, and prior reviewer feedback. - Executor: uses only that
SKILL.mdandskills/<skill>/evals/fixture.mdto produce the real deliverable. - Reviewer: uses
references/approval-rubric.mdto grade the skill and executor output. The filename is legacy; the role is review, not approval.
Stop rules:
- Score >= 90: keep and finalize.
- Five iterations without 90: escalate to the human with the trace.
- Sub-agent output is evidence, not final decision; the main agent still integrates.
Bootstrap:
- This meta-skill is first judged by the main agent against the seed rubric.
- After bootstrap, all other skill rewrites use the tri-agent loop.
Output Format
skill:
name: ""
path: skills/<skill>/SKILL.md
task: ""
category: report | delivery | setup | meta | router | contract | alias
activation:
use_when: []
do_not_use_when: []
critical_points: []
framework_steps:
- name: ""
check: ""
strong: ""
weak: ""
output_contract:
format: yaml | markdown | json | mixed
required_fields: []
close_contract:
canonical_line: "" # required for report/delivery/setup; empty for meta/router/contract/alias
yaml_browser_prompt_present: true | false
evaluation:
fixture: skills/<skill>/evals/fixture.md
rubric: skills/seo-skills-creator/references/approval-rubric.md
threshold: 90
status: draft | kept | escalatedExamples
Example: Strong Skill Direction
Input: "Rewrite seo-analysis."
Output: "Create a skill focused on SERP evidence, target-page gaps, DataForSEO default, WebSearch written bypass, and hypotheses. Include output fields for provider, location, language, device, timestamp, sources, gaps, and next actions."
Example: Weak Skill Direction
Input: "Rewrite seo-analysis."
Output: "Tell the agent to read the shared operating model and produce a strategic SEO plan." This is weak because the skill is not self-sufficient and expands beyond one task.
Related Skills
seo-tools-creator: use for provider CLIs, integrations, registries, and attribution.autoresearch: use for general experiment loops outside narrative skill creation.
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.

