Context
The Neo Agent OS uses a Progressive Disclosure pattern for agent skills — on-demand loaded instruction sets that minimize system prompt token bloat while giving agents specialized procedural knowledge. Seven skills exist today, but there is no documentation explaining:
- What the skill pattern is and why it exists
- How the three lifecycle skills compose into a workflow
- How to create new skills
- The token economics behind lazy loading vs. system prompt injection
Proposed Guide: learn/agentos/ProgressiveDisclosureSkills.md
Content Scope
- The Progressive Disclosure Pattern — YAML frontmatter contract, folder structure (
SKILL.md, references/, assets/), trigger-based activation
- Token Economics — Why skills exist: system prompt budget. Loading a 4K-token workflow guide on-demand vs. bloating every session with all 7 skill payloads
- The Lifecycle Triad — How ticket-intake → pull-request → pr-review compose:
- ticket-intake: Pre-Execution Reflection Gate (validation sweep, ROI calculation, rejection protocol)
- pull-request: Post-Implementation Gate ("Stepping Back" reflection, branch mandate, Conventional Commits, state handoff)
- pr-review: Quality Gate (evaluation metrics, graph ingestion tags, LGTM/Required Actions)
- Tactical Skills — neural-link (live inspection), unit-test (Playwright integration)
- The Meta-Skill —
create-skill bootstraps new skills with correct structure
- How Skills Compose with AGENTS.md — Skills are invoked by mandate rules in the root system prompt. The system prompt says when to invoke; the skill says how to execute.
- Adding New Skills — Step-by-step via the
create-skill skill
Skill Inventory
| Skill |
Type |
Purpose |
ticket-intake |
Lifecycle |
Pre-execution validation gate |
pull-request |
Lifecycle |
Post-implementation reflection + PR creation |
pr-review |
Lifecycle |
Structured quality evaluation |
neural-link |
Tactical |
Live application inspection sequences |
unit-test |
Tactical |
Playwright test authoring patterns |
ideation-sandbox |
Creative |
GitHub Discussion brainstorming |
create-skill |
Meta |
Skill authoring guide |
Registration
- Add to
learn/tree.json under Agent OS
- Cross-link from Swarm Intelligence (execution phase)
- Cross-link from Strategic Workflows (skill invocation patterns)
Acceptance Criteria
A2A Context
- Parent: #9981 (Architecture formalization epic)
- Sibling: #9983 (Swarm Intelligence — completed, PR #9984)
- Sibling: Dream Pipeline ticket (created in same session)
- Origin Session:
70334eab-72c9-44a6-8f48-0b6a96604f49
Context
The Neo Agent OS uses a Progressive Disclosure pattern for agent skills — on-demand loaded instruction sets that minimize system prompt token bloat while giving agents specialized procedural knowledge. Seven skills exist today, but there is no documentation explaining:
Proposed Guide:
learn/agentos/ProgressiveDisclosureSkills.mdContent Scope
SKILL.md,references/,assets/), trigger-based activationcreate-skillbootstraps new skills with correct structurecreate-skillskillSkill Inventory
ticket-intakepull-requestpr-reviewneural-linkunit-testideation-sandboxcreate-skillRegistration
learn/tree.jsonunder Agent OSAcceptance Criteria
learn/agentos/ProgressiveDisclosureSkills.mdtree.jsonA2A Context
70334eab-72c9-44a6-8f48-0b6a96604f49