Capability gap analysis
Skill wde123sadw/agent-preflight/skills/capability-gap-analysis
Determine whether an agent already has the capabilities required for a task before recommending or installing anything new. Use when a task may require unfamiliar file formats, external services, browsers, desktop apps, media generation, databases, deployment access, specialized runtimes, large-context handling, or verification tools. Produce a capability map and hand only proven gaps to tool discovery.From its SKILL.md
npx -y skills add wde123sadw/agent-preflight --skill capability-gap-analysisAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
3.3 KB, 596 tokens by cl100k_base, as published. Nobody here has run it
Capability Gap Analysis
Goal
Translate the approved outcome into required capabilities, inventory what is already available, and identify the smallest real gap. Prefer reuse over acquisition.
Define required capabilities
Describe capabilities without product names. Separate:
- required: the outcome cannot be completed or verified without it;
- preferred: improves quality, speed, or cost but has a viable fallback;
- optional: convenience only.
Read capability-taxonomy.md when the task spans multiple systems or artifact types.
Inventory locally first
Use in-scope read-only inspection to check:
- built-in agent tools and active connectors;
- installed skills, plugins, and MCP servers;
- project scripts, dependencies, task runners, and documented commands;
- operating-system CLIs and runtimes;
- authenticated services already placed in scope;
- existing data access, renderers, validators, and test tools;
- native application automation or official APIs.
Do not expose secrets while inspecting configuration. Record capability presence and evidence, not secret values.
Test fitness, not existence
A tool is available only when it can satisfy the actual requirement. Check:
- supported input and output formats;
- required version or platform;
- read versus write capability;
- authentication and permission scope;
- expected quality and reliability;
- verification or preview support;
- cost, quota, latency, and offline requirements;
- whether use would cross a user approval boundary.
Do not mark “browser available” as sufficient when the task requires an existing signed-in Chrome session and only an isolated browser exists.
Choose reuse or gap
For each requirement, label it:
READY: available and fit;READY_WITH_LIMITS: usable with named constraints;FALLBACK: achievable through a less ideal existing path;GAP: no fit capability is available;GATED: capability exists but requires approval, credentials, payment, or external effect.
Do not search externally for READY or optional capabilities.
Emit a capability map
CAPABILITY MAP
Outcome: <approved outcome>
Requirement | Priority | Status | Evidence | Constraint
<capability> | required | READY/GAP/... | <tool/file/check> | <limit>
Smallest proven gaps:
- <capability stated without a product name>
Fallbacks:
- <existing path and tradeoff>
Approval gates:
- <existing capability that still requires permission>
Hand each required GAP to $tool-discovery. Hand each GATED capability or any proposed acquisition to $tool-review-gate.
Stop conditions
Stop when every required capability is READY, READY_WITH_LIMITS, FALLBACK accepted by the user, GAP ready for discovery, or GATED ready for review.
Do not install, enable, authenticate, or modify configuration in this skill.
What ships with it: 2 files
2.1 KB alongside SKILL.md
agents/
- openai.yaml213 B
references/
- capability-taxonomy.md1.9 KB