context-budget

name	context-budget
description	Keep agent context small and high-signal by routing into AGENTS, the nearest docs, and the narrowest correct test lanes.
origin	ECC v1.10.0 adapted for OMERO Docker Extended
upstream	third_party/ecc-v1.10.0/skills/context-budget/SKILL.md

Context Budget

Use this skill when a task is getting broad, slow, or repetitive, or when the user explicitly wants lower token usage.

Upstream baseline

Start from third_party/ecc-v1.10.0/skills/context-budget/SKILL.md for the generic context-audit mindset.

Fast route

• Read AGENTS.md first, then docs/reference/ai-agent-context-routing.md. Do not dump the whole docs tree into context.
• Search the narrow file set with rg before opening files.
• Open at most 4 task-specific files in the first pass: one domain doc, one implementation file, one nearest test module or suite, and one matching skill.
• Run at most 2 refine loops before broadening scope.
• Add at most 3 more files per escalation round: one additional domain doc, one adjacent implementation file, and one more confirming test module.
• If you have opened 8 task-specific files without naming the edit target and verification lane, stop and summarize before reading more.

Repo rules

• Keep agent context small and high-signal.
• Follow the routing doc's numeric caps; they are CI-validated by python3 tools/lint_docs_structure.py.
• Load one domain doc, one nearest test module, and one split verification lane before broadening.
• Summarize long docs once and reuse the summary instead of reopening the same file repeatedly.
• Prefer the nearest skill in .agents/skills/ over re-deriving a workflow from scratch.
• For verification, run only the relevant split test lanes, not every suite by default.