Skip to content
Back to skills

Capability Builder

ASecurity

TAD-native project-owned Agent Skill creation via create with behavioral proof and safe harness-neutral materialization.

  • 3 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 6, 2026
ai-agents

Security analysis

A100/100

Pro scans all 4 files and shows the line behind each finding

Scanned October 6, 2026

npx -y skills add Sheldon-92/TAD --skill capability-builder --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Capability Builder?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Capability Builder
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/sheldon-92-capability-builder/badge)](https://www.skillsdirectory.com/skills/sheldon-92-capability-builder)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: capability-builder
description: TAD-native project-owned Agent Skill creation via create with behavioral proof and safe harness-neutral materialization.
---

# $capability-builder — Project-Owned Agent Skill Builder (Phase 1 Create + Phase 2 Evolve + Phase 3 Package)

Use for creating exactly one project-owned Agent Skill that is validated, behavior-proven, and safely materialized into `.agents/skills/<name>/`; for signal-driven evolution; and for explicit one-Skill/one-Plugin packaging.

## Ownership Directions

- **Framework Skills (this Builder, alex, blake, gate):** editable authority `.agents/skills/` (sole source since v3.0.0; the Claude mirror direction was removed).
- **Builder-created downstream Skills:** editable authority `.agents/skills/<name>/` via direct materialization (v3.0.0: the `project` projection was removed).

## Router — Phase Selection

On activation, detect intent:

- `$capability-builder create` — Phase 1 (this handoff). Valid. Execute create protocol below.
- `$capability-builder evolve` — Phase 2. Valid. You MUST read `references/evolve-protocol.md` before any design or file work, then follow the protocol.
- `$capability-builder package` — Phase 3. Valid. You MUST read `references/package-openai-plugin.md` before any design or file work, then follow the protocol.

If no explicit phase is given, ask via AskUserQuestion: `create | evolve | package`, then route.

## Create — Mandatory Protocol Load (non-circular trigger)

**When `create` is selected, you MUST read `references/create-protocol.md` before any design or file work.** This load is mandatory. The protocol defines Alex/Blake boundaries, state machine, fixture design, materialization order, and Gate transitions. The trigger lives in this body (circular-trigger safe): without this line, `create` would be undiscoverable.

Progressive disclosure:
- `references/create-protocol.md` — always on `create` (state machine + roles + evidence).
- `references/evolve-protocol.md` — always on `evolve` (trigger enum + fixture-first + stops).
- `references/package-openai-plugin.md` — always on `package` (scaffold/validate/install/drift).
- `../capability-upgrade/references/legacy-pack-research.md` — only after a named evidence gap is found (deep research conditional). See protocol for when.

## Role Boundary (Alex vs Blake)

- **Alex:** decides if a Skill is justified, selects minimal structure, defines discriminative fixture + CONTROL/WITH threshold, writes handoff. Never directly authors a downstream Skill while active.
- **Blake:** authors canonical `.agents/skills/<name>/SKILL.md`, captures CONTROL/WITH with identical prompt, runs `capability-skill.sh validate` and eval, projects only after `BEHAVIOR_PROVEN`, verifies byte identity, completes Ralph Loop, supplies Gate 3 evidence.
- **Gates:** Gate 2 before handoff, Gate 3 after Ralph Loop, Gate 4 human acceptance.

## Out-of-Scope Stops

Return `STOPPED_WITH_REASON` (no writes) when:
- one-off task with no plausible reuse;
- behavior cannot be stated as discriminative fixture;
- duplicates existing Skill without concrete gap;
- requests TAD core change, legacy pack retirement, Plugin/marketplace work, catalog, or DSH.

Return `LEGACY_PACK_OUT_OF_SCOPE` for existing Capability Pack maintenance (route to `capability-upgrade` compatibility path).

## Entry Conditions

`create` may start from:
1. explicit human request with project-local materials; or
2. one accepted SCAND plus linked evidence.

## Minimal Structure (adaptive)

```
.agents/skills/<name>/
└── SKILL.md                 # always required
    references/              # only when knowledge too large/conditional
    scripts/                 # only when deterministic execution required
    assets/                  # only when Skill consumes stable assets
```

No empty directories. No speculative files. No `CAPABILITY.md`, root `README.md`/`CHANGELOG.md`, or per-Skill `install.sh`.

## References

- `references/create-protocol.md` — create state machine and evidence flow (mandatory on create).
- `references/evolve-protocol.md` — evolve state machine and evidence flow (mandatory on evolve).
- `references/package-openai-plugin.md` — package scaffold/validate/install/drift (mandatory on package).

Files in this skill

  • SKILL.md4.2 KB
  • references/create-protocol.md5.6 KB
  • references/evolve-protocol.md3.4 KB
  • references/package-openai-plugin.md3.3 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…