| name | record-triage |
| description | Classify synthetic records with public rules and route ambiguous cases for review. Use when demonstrating reusable record-triage guidance. |
Capability Module Starter
Template Status
This checked-in file is a framework-neutral starter for a capability module.
At templates/SKILL.md it is not itself an Agent Skills package because its
parent directory does not match name. For open-standard compatibility, copy
it into a matching directory, for example:
skills/record-triage/SKILL.md
Other capability modules can use capabilities/<name>/module.md, hosted
instructions, reusable subgraphs, or other framework-specific structures.
Standard Frontmatter
The Agent Skills standard requires name and description. name must be
1–64 characters, use lowercase letters, digits, and single hyphens, have no
leading or trailing hyphen, and match its parent directory. description must
be 1–1024 characters and explain what the skill does and when to use it.
Optional standard fields are license, compatibility, metadata, and the
experimental allowed-tools. compatibility is 1–500 characters when used;
metadata maps string keys to string values. allowed-tools is a string whose
support and authority semantics vary by implementation. A declaration never
creates universal authorization; apply runtime permissions and governance.
Do not add vendor-only fields to a portable default. See the
Agent Skills mapping guide.
When to Use
Use this module when a fabricated record must be classified with published
synthetic rules. Do not use it for real or consequential records.
Inputs
- Synthetic record identifier
- Fabricated summary
- Approved public category rules
Workflow
- Read the applicable rule from
references/category-rules.md only when the
task requires it.
- Compare the fabricated record with each eligible rule.
- Apply one category only when exactly one rule matches.
- Route missing, conflicting, or ambiguous evidence for human review.
- Format metadata using
assets/triage-result-template.json when needed.
Outputs and Failure Behavior
Return a synthetic identifier, category or needs-review, applied rule,
rationale, and review flag. Stop rather than guess when inputs are missing,
private, real, out of scope, or ambiguous.
Progressive Disclosure
Agents can use name and description for discovery, load the complete body
when activating the package, and load scripts, references, assets, or other
resources only as needed. Keep SKILL.md concise and resolve supporting paths
from the skill root.
Skill and Tool Boundary
A capability module or skill packages reusable task guidance. A tool is a
callable action with an interface, inputs, outputs, permissions, side effects,
and failure behavior. Instructions about a tool—or executable helpers in an
optional scripts/ directory—do not make the whole skill package a tool.
Supply-Chain and Public-Safety Review
Before importing or activating a package, review:
- provenance, license, source, version, and update history;
- instructions, scripts, references, assets, and dependencies;
- network needs, tool declarations, file and environment access;
- permissions, side effects, failure behavior, validation, and eval evidence.
Use only generic synthetic examples. Do not include secrets, private records,
proprietary procedures, raw logs, memory, state, or unsafe automation.
Structural conformance does not establish safety or authorization.
