Context
The ideation-sandbox skill currently contains the entire heavy payload (rules, annotations, taxonomy, mindsets) within the root SKILL.md file. As the skill has evolved to include detailed mandates (e.g., the #10119 annotation pattern, PR depth challenges, OQ resolution tags), the file has grown too large for efficient context assembly.
The Problem
When Neo.ai.context.Assembler parses the skill, including the full heavy payload inside SKILL.md bloats the system prompt. According to the Anthropic Progressive Disclosure strategy outlined in the create-skill meta-skill, SKILL.md must be a "stupidly short" lightweight router, and all heavy instructional payloads must be deferred to the references/ directory.
The Architectural Reality
The skill architecture dictates a clear separation of concerns:
.agent/skills/<skill>/SKILL.md -> Lightweight Router (Trigger conditions, high-level summary, explicit file pointers)..agent/skills/<skill>/references/<doc>.md -> Heavy Payload (Deep logic, annotation patterns, calibration rubrics).
The Fix
- Create
.agent/skills/ideation-sandbox/references/ideation-sandbox-reference.md.
- Move all detailed workflow rules, taxonomy, mindset shifts, and the annotation pattern from
.agent/skills/ideation-sandbox/SKILL.md into the new reference file.
- Rewrite
.agent/skills/ideation-sandbox/SKILL.md to act strictly as a router, maintaining only the YAML frontmatter and brief instructions pointing agents to use view_file on the reference document when triggered.
Acceptance Criteria
Out of Scope
- Modifying the underlying logic or rules of the ideation-sandbox workflow (this is purely structural).
Related
- PR #10280 (feat: formalize ideation-sandbox iterative workflow)
- Discussion #10119 (Target for ideation calibration)
Origin Session ID: 0b29a8fa-c6b0-42e2-ab3b-8015a99db2d8
Retrieval Hint: "ideation-sandbox progressive disclosure refactor"
Context
The
ideation-sandboxskill currently contains the entire heavy payload (rules, annotations, taxonomy, mindsets) within the rootSKILL.mdfile. As the skill has evolved to include detailed mandates (e.g., the#10119annotation pattern, PR depth challenges, OQ resolution tags), the file has grown too large for efficient context assembly.The Problem
When
Neo.ai.context.Assemblerparses the skill, including the full heavy payload insideSKILL.mdbloats the system prompt. According to the Anthropic Progressive Disclosure strategy outlined in thecreate-skillmeta-skill,SKILL.mdmust be a "stupidly short" lightweight router, and all heavy instructional payloads must be deferred to thereferences/directory.The Architectural Reality
The skill architecture dictates a clear separation of concerns:
.agent/skills/<skill>/SKILL.md-> Lightweight Router (Trigger conditions, high-level summary, explicit file pointers)..agent/skills/<skill>/references/<doc>.md-> Heavy Payload (Deep logic, annotation patterns, calibration rubrics).The Fix
.agent/skills/ideation-sandbox/references/ideation-sandbox-reference.md..agent/skills/ideation-sandbox/SKILL.mdinto the new reference file..agent/skills/ideation-sandbox/SKILL.mdto act strictly as a router, maintaining only the YAML frontmatter and brief instructions pointing agents to useview_fileon the reference document when triggered.Acceptance Criteria
ideation-sandbox/SKILL.mdis reduced to a lightweight router.ideation-sandbox/references/ideation-sandbox-reference.md.pull-request,pr-review) correctly point to the new reference location if applicable.Out of Scope
Related
Origin Session ID: 0b29a8fa-c6b0-42e2-ab3b-8015a99db2d8 Retrieval Hint: "ideation-sandbox progressive disclosure refactor"