Task effort-tier convention

← All Specs

Task effort-tier convention

Effort: quick

Goal

Document the unified effort-tier vocabulary in SciDEX/AGENTS.md so quest generators, watchdog tasks, and human spec authors all use the same names. Pairs with [Orchestra #209](https://github.com/SciDEX-AI/Orchestra/pull/209) which provides the runtime mapping.

Acceptance Criteria

☑ AGENTS.md "Task Spec Requirement" section lists Effort as a required spec line.
☑ New "Task Effort Tier" subsection documents the 5 tiers (quick/standard/deep/thorough/extensive) with use cases and the claude/codex level mapping.
☑ Example spec in the doc includes the Effort: line.
☑ Mention of --effort CLI flag.

Approach

  • Edit AGENTS.md after the existing "Task Spec Requirement" section.
  • Cross-link Orchestra/AGENTS.md §5b for the canonical runtime spec.
  • Add SciDEX-specific guidance: which task families benefit from extensive (challenge creation, hypothesis generation) vs quick (linking, scoring with rubric).
  • Work Log

    2026-04-27 — Added effort-tier section

    • Documented the 5-tier vocabulary
    • Added required Effort: spec line
    • Cross-linked Orchestra runtime mapping

    File: task_effort_tier_convention_spec.md
    Modified: 2026-05-01 20:13
    Size: 1.2 KB