Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
An effective AI agent skill is a small, versioned directory—not a giant prompt. Put a required SKILL.md manifest at its root, use front matter to say exactly what the workflow does and when it should activate, then put the executable procedure in the body. Move policy, background, deterministic scripts, templates, and other large material into linked references/, scripts/, and assets/ directories.
The biggest reduction in guesswork comes from making every decision visible: required inputs, exact outputs, allowed tools and files, branch conditions, facts the agent must not infer, stopping conditions, and requests that should not trigger the skill. This guide shows how to design, package, test, and maintain that structure across common agent hosts.
As an Amazon Associate I earn from qualifying purchases.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11What is an AI agent skill file, and why does it reduce guesswork?
A skill is a folder of instructions and resources that teaches an agent a repeatable workflow. The folder is anchored by a required SKILL.md file. Its front matter is used for discovery and routing; its instruction body explains how to execute the workflow; optional linked files supply detail or machinery only when needed.
This arrangement solves two different problems. Metadata helps the host decide whether a request matches the skill. The body then tells the agent what to do after activation. If either part is vague, the agent has to fill gaps from general knowledge, which produces inconsistent tool choices, missing steps, or invented values.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
Documentation from OpenAI, Anthropic, and other implementations describes this as progressive disclosure:
- Discovery layer: the host sees the skill name and description first.
- Execution layer: the host loads the complete
SKILL.mdbody when the request appears relevant. - Supporting layer: the agent reads a linked reference, runs a script, or copies an asset only when the procedure tells it to.
Do not treat a skill as a guarantee that an agent will always behave correctly. The published guidance is normative rather than a quantitative benchmark, so there is no universal percentage improvement to promise. Treat explicit routing and validation as engineering controls, then test them with the prompts your users actually send.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsHow should I structure a skill directory?
Keep the root predictable and make every dependency discoverable from the manifest. A minimal directory contains only SKILL.md; a production skill can add role-specific subdirectories.
invoice-reconciliation/
├── SKILL.md
├── references/
│ ├── accounting-policy.md
│ └── field-map.md
├── scripts/
│ ├── validate_csv.py
│ └── totals.py
└── assets/
└── reconciliation-report.xlsxThe names of the optional directories are conventions, not a license to hide information. Link each file from the body and explain when it should be loaded. An unreferenced file is effectively invisible to a model and may be omitted from a future package.
Use one directory for one coherent job. If invoice reconciliation and customer-email drafting have different inputs, permissions, failure modes, and outputs, make two skills rather than one broad “finance assistant” skill. Narrow scope makes both routing and testing easier.
How do I choose the right scope for one skill?
Start with the artifact or decision the user needs, not with a department name or a collection of vaguely related capabilities. A good scope can be expressed in one sentence: “Convert a validated CSV of invoices into a reconciliation report,” for example. A poor scope sounds like “Handle finance tasks,” because it gives the router no reliable boundary.
Free tools Windows power users keep installed
One-click scans. No signup required.
Define the one job
- Name the initiating request in user language.
- Identify the single workflow that follows.
- Specify the final artifact, message, or state change.
- List adjacent requests that belong to another skill.
Split a skill when two branches need different permissions, tools, or evidence. Combining them may save a directory but forces the agent to choose among unrelated procedures before it has enough information to choose safely.
Write a non-trigger boundary
Include a clear “do not use when” condition in the description or body. For example, a skill that reconciles existing invoices should not activate for tax advice, vendor onboarding, or free-form bookkeeping questions. A negative boundary is as important as a positive trigger because it prevents accidental invocation on similar words.
What belongs in SKILL.md front matter?
Front matter should be short, valid for the host, and sufficient for discovery. Anthropic’s platform documentation requires name and description; its naming guidance uses lowercase letters, numbers, and hyphens. Other hosts may add fields, so follow the target host’s schema instead of assuming that every implementation accepts the same keys.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
---
name: invoice-reconciliation
description: Reconcile a validated invoice CSV against payment records and produce a report with exceptions. Use when the user asks to match invoices, payments, or outstanding balances. Do not use for tax advice, accounting policy interpretation, or vendor onboarding.
---The description should state both capability and trigger conditions. OpenAI guidance recommends writing descriptions around the user goal and the conditions that should activate the workflow. Avoid marketing language, implementation trivia, and a long list of every possible synonym; the router needs a compact decision rule.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Make the name stable
Use a short, lowercase, hyphenated identifier that will remain useful if the prose changes. Do not encode a temporary model name, a personal name, or an unpinned date in the identifier. If the host has stricter naming rules, validate the name before packaging.
Keep metadata separate from procedure
Do not put a ten-step process in the description. Metadata is loaded early and often; excessive detail increases routing noise and consumes context before the workflow is known to be relevant. Put execution details in the body and supporting files.
How do I specify inputs and outputs so the agent cannot guess?
State the contract before the procedure. Required inputs should include files, parameters, permissions, and acceptable formats. Outputs should describe the exact artifact or response, its location or format, and what to report when the workflow cannot complete.
Inputs checklist
- Required files and their formats, such as UTF-8 CSV with named columns.
- Required parameters, including date range, currency, locale, or account identifier.
- Permissions the agent must have before reading or writing anything.
- Validation rules, such as non-empty identifiers and unique row keys.
- What to do when an input is absent, malformed, or contradictory.
Outputs checklist
- The exact file, message, or system change to produce.
- Required fields, ordering, naming, and destination.
- Whether a human approval is required before delivery.
- The error format when a result cannot be safely produced.
- Whether intermediate files must be retained or deleted.
Use observable language. “Return a useful report” leaves too much room for interpretation; “write reconciliation_report.xlsx with Summary, Matched, Exceptions, and Unmatched sheets” gives the agent a testable target.
How should I write the procedure inside SKILL.md?
Write imperative steps in execution order. Every branch should name its condition and the next action. Every irreversible action should have an explicit approval or verification point.
- Parse and validate inputs. Check that required files exist, permissions are available, columns have the expected names, and values meet the stated constraints.
- Stop on missing prerequisites. Report the specific missing item and ask only for that item. Do not invent a filename, date, account, or permission.
- Choose the branch. Use the documented condition, such as file type, record count, or whether an approval flag is present. If more than one branch applies, follow the stated priority order.
- Load supporting material selectively. Read a policy reference for a policy decision, run a validation script for deterministic checks, or copy an asset for a required template. Do not load every file by default.
- Perform the transformation. Follow the listed commands, field mappings, and ordering. Preserve source values unless the procedure explicitly allows normalization.
- Validate the result. Check totals, required fields, file integrity, and any business-rule assertions before claiming success.
- Deliver or stop. Write the output to the specified destination only after validation. If a check fails, identify the failed condition, preserve evidence, and stop rather than silently repairing an ambiguous value.
Show branches explicitly
Use a compact pattern that an agent can execute:
If the input has no header row:
1. Stop.
2. Tell the user that a header row is required.
3. Do not infer column names.
If duplicate invoice IDs are present:
1. Run scripts/validate_csv.py.
2. Read references/field-map.md only if the script reports a known alias.
3. If duplicates remain, produce an exceptions report and request confirmation.This is safer than a sentence such as “handle duplicates appropriately,” which gives no criterion for choosing a repair.
Which boundaries and stopping conditions should I include?
Boundaries prevent an agent from turning uncertainty into fabricated certainty. OpenAI’s build guidance specifically recommends stating what the model must not infer, which tools or files it may use, and when it must stop or ask for clarification.
- No inference: do not guess missing dates, currencies, column meanings, identities, or permissions.
- Tool boundary: use only the named tools and scripts; do not substitute an unapproved external service.
- File boundary: read and write only the paths listed by the procedure.
- Authority boundary: do not approve payments, publish content, or change production data without the stated human approval.
- Evidence boundary: distinguish a value found in an input from a value calculated by the workflow.
- Stop condition: halt on failed validation, conflicting records, missing consent, or an unavailable required dependency.
Say what the agent should report at a stop: the condition, the affected item, the evidence available, and the smallest action the user can take to continue. “Cannot proceed” without that detail simply moves the guesswork to the user.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
How can examples improve routing and edge-case handling?
Examples are executable routing tests. Include several positive prompts that should activate the skill, at least one negative prompt that should not, and an edge case that distinguishes a documented branch.
Positive examples
- “Match this invoice CSV to the payment export and create the exceptions workbook.”
- “Reconcile the attached payment records for April and flag unmatched invoice IDs.”
Negative example
- “Explain how sales tax works for invoices in another country.” This should not activate an operational reconciliation skill because it asks for general tax guidance.
Edge-case example
- “The payment export has two currencies and no exchange-rate column.” The skill should stop and request the approved conversion source instead of choosing a rate.
Examples should use the language real users use, including abbreviations and incomplete requests. They should not contradict the formal input contract. If a prompt can reasonably match two skills, revise the descriptions or add a routing rule rather than hoping the model will infer priority.
Where do references, scripts, and assets belong?
Separate supporting material by role so the agent knows why it is loading a file. This also keeps the main manifest short enough for progressive disclosure.
| Directory | Use it for | Good contents | Instruction in SKILL.md |
|---|---|---|---|
references/ |
Policy and background that require interpretation | Field definitions, compliance rules, decision tables, domain terminology | “Read references/accounting-policy.md when a policy exception is encountered.” |
scripts/ |
Deterministic helpers and validation | Parsers, format checks, calculators, report generators | “Run scripts/validate_csv.py before transforming rows; treat a nonzero exit as a stop.” |
assets/ |
Reusable files copied or filled during execution | Templates, schemas, sample configuration, report workbooks | “Copy assets/reconciliation-report.xlsx before writing the final sheets.” |
Do not hide procedural rules in a script’s comments or a template’s formatting. The body should still state the decision logic, expected inputs, and success criteria. Scripts can enforce deterministic checks, but they cannot replace an explicit permission or approval policy.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →How does progressive disclosure affect the writing?
Progressive disclosure is a context strategy, not merely a folder layout. Keep the metadata sufficient for routing, keep the body self-contained for the normal path, and defer bulky or specialized material until a branch requires it.
Level one: metadata
Use a concise name and description. A host may inspect this layer across many installed skills, so avoid unnecessary prose and volatile implementation details.
Level two: the main procedure
Include the normal workflow, validation, boundaries, and links to supporting files. A reader should be able to understand the contract without opening every reference.
Level three: linked resources
Load only the policy, script, or asset relevant to the current branch. Explain the loading condition next to the link. This reduces context consumption and lowers the chance that unrelated instructions will be mistaken for the active procedure.
Recommended Free Tools
When a reference is large, split it by decision or domain rather than creating one encyclopedic file. When a script changes behavior, update its invocation and expected exit conditions in SKILL.md at the same time.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
What differs between OpenAI/Codex, Claude, VS Code, and other hosts?
The core pattern is portable, but discovery and execution details are host-specific. Confirm the target implementation’s directory locations, naming rules, implicit-invocation behavior, and tool permissions before distributing a skill.
| Comparison axis | Questions to answer | Why it matters |
|---|---|---|
| Host compatibility | Does the host support this skill format, linked files, and scripts? | A valid folder in one product may be ignored by another. |
| Discovery rules | Where are project and personal skills installed? Is invocation implicit, explicit, or both? | The same description may route differently depending on the host. |
| Context strategy | Which metadata is preloaded, and when are linked resources read? | Large front matter can waste context; hidden links can remain undiscovered. |
| Execution needs | Does the workflow require shell commands, code execution, file writes, or only prose? | Tool permissions must be available before the skill can complete. |
| Maintenance | Can the directory, scripts, and assets be versioned and reproduced together? | A prose update without its helper code can create silent drift. |
OpenAI’s API guidance treats a skill as a directory anchored by a manifest with front matter and instructions. Anthropic’s documentation describes the same three-level structure and requires front matter fields named name and description. VS Code documentation describes project and personal skill locations and states that SKILL.md defines behavior. These statements establish a common design pattern, not identical installation commands; use the host’s current documentation for the final path and loader behavior.
How do I test whether a skill routes reliably?
Test routing and execution separately. A skill can have an excellent procedure but never activate, or activate correctly but fail because its inputs and tools are underspecified.
- Create a prompt set containing clear positives, paraphrased positives, incomplete requests, near-matches, and explicit negatives.
- Check whether the intended skill activates for each positive and stays quiet for each negative.
- For every activation, verify that the agent asks for missing prerequisites instead of inventing them.
- Exercise every documented branch, including malformed files, duplicate records, conflicting values, and unavailable tools.
- Compare the produced artifact with the output contract: filename, format, fields, calculations, and destination.
- Repeat the tests after changing the description, procedure, scripts, references, or assets.
Use a release checklist
- The directory contains exactly one required root manifest named
SKILL.md. - Front matter parses and follows the target host’s naming rules.
- The description states capability, positive triggers, and at least one non-trigger condition.
- Every required input, output, permission, and stopping condition is explicit.
- Every linked file exists at the stated relative path.
- Commands specify their working directory, arguments, expected output, and failure behavior.
- Examples include positive, negative, and edge-case prompts.
- The package records a version or commit identifier for reproducibility.
What are the common failure modes and fixes?
Most “guessing” problems are specification problems. Use the symptom to locate the missing contract.
| Symptom | Likely cause | Fix |
|---|---|---|
| The skill activates for unrelated requests. | The description names a broad topic rather than a specific workflow. | Replace topic words with the user goal, required input, and a “do not use when” condition. |
| The skill never activates. | Trigger language does not match user phrasing, or the skill is installed outside the host’s discovery path. | Add realistic paraphrases to tests and verify the host’s project or personal location. |
| The agent invents a missing value. | The procedure says to “complete” or “normalize” data without a no-inference rule. | State the exact stop condition and the question to ask for the missing value. |
| A reference file is ignored. | The file is present but not linked or its loading condition is unclear. | Add a relative link and a branch that names when to read it. |
| A script runs but the result is trusted blindly. | Exit codes, output fields, or failure handling are not documented. | Define expected output and tell the agent to stop on a nonzero exit or failed assertion. |
| Different hosts behave differently. | Directory locations, implicit invocation, or tool permissions differ. | Maintain host-specific packaging notes and test each supported host separately. |
| An update breaks an old workflow. | Scripts, assets, and prose changed without a reproducible version. | Pin a release identifier, review bundled code with prose changes, and retain a rollback point. |
What should a complete SKILL.md look like?
The following compact example demonstrates the essential pieces without pretending that one template fits every host.
---
name: invoice-reconciliation
description: Reconcile a validated invoice CSV against payment records and produce a report with exceptions. Use when the user asks to match invoices, payments, or outstanding balances. Do not use for tax advice, accounting policy interpretation, or vendor onboarding.
---
# Skill purpose
Reconcile invoice rows to payment rows and create a reviewable workbook.
## Inputs
- UTF-8 invoice CSV with a unique invoice_id column.
- Payment CSV with invoice_id, amount, currency, and paid_at columns.
- A reporting period supplied by the user.
- Permission to read the input files and write the output directory.
## Outputs
- reconciliation_report.xlsx with Summary, Matched, Exceptions, and Unmatched sheets.
- A short response stating the output path and any unresolved exceptions.
## Procedure
1. Confirm that both CSV files exist and that the reporting period is supplied.
2. Run scripts/validate_csv.py on each file. Stop on a nonzero exit code.
3. Read references/field-map.md only when a documented column alias is reported.
4. Match rows by invoice_id, then compare amount and currency.
5. If duplicate IDs, missing currencies, or conflicting amounts remain, write them to Exceptions and stop before approval.
6. Copy assets/reconciliation-report.xlsx and populate the four required sheets.
7. Run scripts/totals.py and verify that matched plus unmatched totals equal the source total.
8. Return the workbook path and list unresolved exceptions. Do not invent exchange rates or invoice IDs.
## Boundaries
- Do not infer a reporting period, currency, exchange rate, or column meaning.
- Do not change source CSV files.
- Do not approve payments or publish the workbook without human approval.
## Examples
- Use when: “Match these invoices to the payment export.”
- Use when: “Create an exceptions workbook for April payments.”
- Do not use when: “Explain sales-tax obligations for a foreign invoice.”
- Edge case: two currencies without an approved rate requires a stop and a request for the rate source.
## Supporting files
- references/field-map.md: read only for documented column aliases.
- scripts/validate_csv.py: run before transformation; nonzero exit stops the workflow.
- scripts/totals.py: run before delivery to verify totals.
- assets/reconciliation-report.xlsx: copy as the output template.Notice what the example does not do: it does not promise that the agent will resolve every discrepancy, silently choose an exchange rate, or use unlisted tools. Those omissions are deliberate safety boundaries.
How should I version and maintain a skill?
Package the manifest, references, scripts, and assets as one reproducible unit. OpenAI’s API cookbook, dated February 10, 2026, recommends version pinning for production bundles and treating bundled code and resources as carefully as prose.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →- Assign a release identifier or commit to every production package.
- Review changes to scripts and templates alongside instruction changes.
- Record host compatibility and required permissions.
- Retest positive, negative, and edge-case prompts after each change.
- Keep a rollback version when a routing or output change could affect existing users.
- Remove obsolete references rather than leaving conflicting rules in the directory.
Versioning is especially important when a helper script changes a calculation or file format. The agent may follow the old prose while executing new code unless the package is updated and tested as a single release.
What can you realistically expect from a well-written skill?
A precise skill narrows the agent’s decision space: it tells the host when to invoke the workflow, gives the model an ordered procedure, and supplies deterministic helpers or authoritative references at the point of need. It cannot compensate for missing permissions, contradictory source data, or an unsupported host.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
Measure success with operational checks rather than a generic promise: correct activation on intended prompts, refusal on non-triggers, explicit requests for missing information, valid output artifacts, and clean stops on unsafe conditions. Those checks turn “it should know what to do” into behavior you can inspect and improve.
FAQ
Can a skill automatically invoke another skill?
Do not assume that one skill can call another merely because both are installed. If a workflow depends on a second skill, document the handoff, required inputs, and expected result, then verify that the host supports explicit delegation or ask the user to invoke the next skill.
Should every skill include executable scripts?
No. Instruction-only workflows are valid when the task is explanatory or editorial. Add a script when a deterministic check, parser, calculation, or file transformation is safer and more repeatable than asking the model to perform it in prose.
How long should the description be?
There is no universal character limit across hosts. Make it short enough for early routing while still naming the capability, the user requests that should trigger it, and a meaningful non-trigger; move examples and procedure details into the body.
Where should secrets and credentials go?
Do not place secrets in SKILL.md, references, assets, or example prompts. Declare the required permission or environment-provided credential, tell the agent what to do when it is unavailable, and follow the host’s secret-management mechanism.
Is a skill file the same as a system prompt?
No. A system prompt usually applies globally, while a skill is a selectively routed package for one repeatable job. A skill can contain instructions that are loaded only for matching requests, plus scripts and files that a single global prompt cannot organize as cleanly.
Frequently Asked Questions
Can a skill automatically invoke another skill?
Do not assume that one skill can call another merely because both are installed. If a workflow depends on a second skill, document the handoff, required inputs, and expected result, then verify that the host supports explicit delegation or ask the user to invoke the next skill.
Should every skill include executable scripts?
No. Instruction-only workflows are valid when the task is explanatory or editorial. Add a script when a deterministic check, parser, calculation, or file transformation is safer and more repeatable than asking the model to perform it in prose.
How long should the description be?
There is no universal character limit across hosts. Make it short enough for early routing while still naming the capability, the user requests that should trigger it, and a meaningful non-trigger; move examples and procedure details into the body.
Where should secrets and credentials go?
Do not place secrets in SKILL.md, references, assets, or example prompts. Declare the required permission or environment-provided credential, tell the agent what to do when it is unavailable, and follow the host’s secret-management mechanism.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Is a skill file the same as a system prompt?
No. A system prompt usually applies globally, while a skill is a selectively routed package for one repeatable job. A skill can contain instructions that are loaded only for matching requests, plus scripts and files that a single global prompt cannot organize as cleanly.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
