Contextual agent skill creation
Turn your task, files and selected Memory into reusable agent skills, with A+ as the instruction-quality target.
Use this release-note briefExplore team workflows →
Illustrative skill brief
Create a SKILL.md skill for reviewing draft release notes against an approved change list. Use it when a teammate asks to prepare or check release notes. Compare each stated feature with the supplied changes, flag unsupported availability claims and return a proposed revision plus unresolved questions. Do not invent product…
Read the full example →
What is the Prompeteer agent skill generator?
Prompeteer creates contextual SKILL.md agent skills from a task description and the context you provide. Individuals can turn recurring work into reusable instructions; teams can document a shared method with explicit inputs, review criteria and handoff boundaries.
Describe when the skill should be used, what material it needs and the output you expect. Add relevant files or selected Memory through the supported context flow. Keep current business facts separate from the reusable method so the next run can use the right sources.
Sources: Prompeteer product overview · Supported connections and Memory access
Aim for an A+ Skill Score, then test the skill
A+ Skill Score is the quality target for agent skills. Skill Score evaluates effectiveness, specificity and description quality. For prompt creation, A+ Prompt Score is the corresponding target, with feedback across 16 prompt-quality dimensions. Use the relevant feedback to improve the instructions and context. A score is an aid to reviewing the instructions, not an independent certification or a promise that every generated skill will receive A+.
Check the resulting skill with representative inputs in your chosen agent. Include a case with complete context and a case with a missing source; a useful review skill should identify the gap instead of inventing the answer.
Source: Skill Score and Prompt Score: quality checks and limits
A skill for release-note review
The instruction should make the review reproducible without silently publishing the result.
Illustrative skill brief
Create a SKILL.md skill for reviewing draft release notes against an approved change list. Use it when a teammate asks to prepare or check release notes. Compare each stated feature with the supplied changes, flag unsupported availability claims and return a proposed revision plus unresolved questions. Do not invent product behavior, modify the change list or publish the notes. Include an example with a missing rollout date.
Inspect a tested SKILL.md example
Internal format test. This authored example uses the release-note brief above. The test runs the displayed text through Prompeteer’s production normalizer and strict skill validator, with no model calls. It checks the file structure and preserved instructions; it does not measure model quality or runtime safety.
From the public brief to a checkable file
- Brief Review release notes against supplied, approved changes.
- Format Normalize an outer Markdown fence and validate the name, description and body.
- Boundary Keep source checks, missing-date handling and the prohibition on publishing intact.
| Check | Recorded fixture result |
|---|---|
| Required metadata | Name and description pass the production strict validator. |
| Workflow boundary | The normalized body retains source checks and does not authorize publication. |
| Negative control | An input without the required frontmatter is rejected when automatic repair is disabled. |
Validator-tested skill example
---
name: release-note-review
description: Review draft release notes against an approved change list. Use when preparing or checking a release announcement.
---
# Release-note review
## Inputs
Ask for the draft release notes and approved change list. If either is missing, request it before reviewing claims.
## Workflow
1. Match every stated feature to an approved change.
2. Flag unsupported product behavior, availability claims and rollout dates.
3. Propose a revision that preserves the supported facts.
4. Return unresolved questions separately from the revised draft.
## Output
Return a claim-to-source table, a proposed revision and unresolved questions. Cite the supplied change-list item for each supported claim.
## Limits
Do not publish the notes, modify source records or invent product behavior. A proposed revision is not approval to release it.
## Example
The approved change says "CSV export added" but the draft promises availability tomorrow. Flag the missing rollout date and remove that promise until the owner supplies an approved date.
You can inspect or reuse this example. Test any resulting skill in your actual client before giving it access to sensitive data or external actions.
Reproducible example
A release-review skill that passes the file checks
Authored release-note skill fixture. This is the actual output of the production Markdown normalizer and strict validator, with automatic repair disabled.
Checked 19 October 2026 · product-workflows-v2 · no model calls
- Input A complete release-review instruction is wrapped in a Markdown code fence.
- File The normalizer removes the outer fence; the validator checks required metadata and body.
- Control A sentence without frontmatter fails, making the result useful beyond the successful example.
- PassedRequired metadata passes strict validation
- PassedPublishing restriction survives normalization
- PassedMissing frontmatter is rejected without repair
Inspect inputs and recorded output
{
"input": {
"brief": "Create a SKILL.md skill for reviewing draft release notes against an approved change list. Use it when a teammate asks to prepare or check release notes. Compare each stated feature with the supplied changes, flag unsupported availability claims and return a proposed revision plus unresolved questions. Do not invent product behavior, modify the change list or publish the notes. Include an example with a missing rollout date.",
"wrappedSkill": "```markdown\n---\nname: release-note-review\ndescription: Review draft release notes against an approved change list. Use when preparing or checking a release announcement.\n---\n\n# Release-note review\n\n## Inputs\nAsk for the draft release notes and approved change list. If either is missing, request it before reviewing claims.\n\n## Workflow\n1. Match every stated feature to an approved change.\n2. Flag unsupported product behavior, availability claims and rollout dates.\n3. Propose a revision that preserves the supported facts.\n4. Return unresolved questions separately from the revised draft.\n\n## Output\nReturn a claim-to-source table, a proposed revision and unresolved questions. Cite the supplied change-list item for each supported claim.\n\n## Limits\nDo not publish the notes, modify source records or invent product behavior. A proposed revision is not approval to release it.\n\n## Example\nThe approved change says \"CSV export added\" but the draft promises availability tomorrow. Flag the missing rollout date and remove that promise until the owner supplies an approved date.\n```",
"autoRepair": false,
"negativeControl": "Review the release notes."
},
"output": {
"skill": "---\nname: release-note-review\ndescription: Review draft release notes against an approved change list. Use when preparing or checking a release announcement.\n---\n\n# Release-note review\n\n## Inputs\nAsk for the draft release notes and approved change list. If either is missing, request it before reviewing claims.\n\n## Workflow\n1. Match every stated feature to an approved change.\n2. Flag unsupported product behavior, availability claims and rollout dates.\n3. Propose a revision that preserves the supported facts.\n4. Return unresolved questions separately from the revised draft.\n\n## Output\nReturn a claim-to-source table, a proposed revision and unresolved questions. Cite the supplied change-list item for each supported claim.\n\n## Limits\nDo not publish the notes, modify source records or invent product behavior. A proposed revision is not approval to release it.\n\n## Example\nThe approved change says \"CSV export added\" but the draft promises availability tomorrow. Flag the missing rollout date and remove that promise until the owner supplies an approved date.",
"valid": true,
"errors": [],
"invalidControlRejected": true,
"invalidControlErrors": [
"Missing YAML frontmatter (must start with --- and end with ---)"
]
}
}
Production functions: normalizeSkillContent, validateSkillContent.
Download the input and result record, or replay all four examples locally with Node.js 20 or later.
Inputs and results (JSON)Offline replay (.mjs)File checksums
node replay-product-workflows-v2.mjs
The replay contains the relevant production functions. It needs no account, credentials, package install or network connection. Inspect the downloaded script before running it.
Format validity does not establish that a client will follow the instructions or that executable code is safe. This example makes no model calls and does not install or execute the skill. Test in your selected client before use.
Source: Agent Skills specification
Create a skill from this reviewed brief
Your next three steps
- Open the release-note brief The button opens Create in Skill mode with this public example. You can edit it before generating.
- Specify the required inputs Keep the approved change list and draft release notes as explicit inputs. Leave missing rollout dates for review.
- Inspect SKILL.md before use Review the trigger, instructions and requested permissions. Follow your target agent’s installation guide; creating a file does not install or run it.
Start from a task or an existing professional workflow
Prompeteer offers a combined library of 5,000+ professional agent workflows and skills for individuals and teams. Browse Skills Hub for a reusable skill, or explore professional workflows to decide which steps your custom skill should repeat. This is the combined catalog size, not the number of individual SKILL.md files.
Choose the smallest workflow that fits, then supply your audience, sources and review criteria. For individual work, keep a tested instruction for the tasks you repeat. For a team, name the person responsible for maintaining the skill, record the source version and agree which actions need a reviewer before reuse.
Sources: Prompeteer Skills Hub · Professional workflows
Portable instructions still need a compatible environment
Inspect the generated files and check the target agent’s installation instructions. A skill can describe a workflow without having permission or tools to execute every action.
The Agent Skills specification defines a shared format. It does not guarantee identical behavior across clients or make executable scripts safe merely because they are packaged in a skill.
Source: Agent Skills specification
Common questions
Is a skill just a long prompt?
A skill packages a repeatable task with metadata, instructions and, when needed, references or other files. The packaging helps compatible agents discover and load the workflow.
What makes a generated skill contextual?
The instruction reflects your task, selected sources, audience and constraints. It identifies which current facts the agent needs and what to do when those facts are missing.
Does A+ guarantee a correct result?
No. A+ refers to a Prompeteer Skill Score quality target for the instructions. Check source accuracy, client compatibility and the actual results separately.
Can I edit a generated skill?
Yes. Review and adapt the generated instructions for your task and client before using them.
Sources and review
Published by Prompeteer. See the people responsible for the product and contact the team with corrections. The linked workflow artifacts describe exactly what was checked; illustrative examples are not customer results.
Reviewed September 17, 2026
- Agent Skills specification
- Prompeteer product overview
- Supported connections and Memory access
- Skill Score and Prompt Score: quality checks and limits
- Prompeteer Skills Hub
- Professional workflows
Continue your workflow
Build agent skills around your context →
Create contextual agent skills from your files, selected Memory and task requirements. Give individual and team workflows current sources and clear review steps.
Find reusable agent skills and review their instructions.
Connect reusable instructions to a repeatable task.
Know what is inside an agent skill →
Read the SKILL.md format, distinguish metadata from instructions and inspect optional scripts and references. Follow the official specification for validation.
Put the workflow to work.
Use this release-note briefExplore team workflows →