AgentSkillsCN

new-ability-low-flow

低层次能力脚手架的分步流程。关键词:能力、低层次、流程。

SKILL.md
--- frontmatter
name: new-ability-low-flow
description: "Step-by-step flow for low-level ability scaffolding. Keywords: ability, low-level, flow."

Flow: New Low-Level Ability

This doc contains the step-by-step flow. For inputs, tools, outputs, and safety, see /.system/skills/ssot/repo/scaffolding/new-ability-low/SKILL.md.


Step-by-Step Flow (AI + Human)

Step 1: Collect Parameters (AI -> Human)

AI asks human for:

  1. Operation key (validate: unique, follows naming convention)
  2. Summary description
  3. Implementation kind (script/api/mcp)
  4. Entrypoints (which ABILITY.md files should list this)
  5. Kind-specific implementation details

Human checkpoint: Confirm parameters before proceeding.

Step 2: Validate Inputs (AI)

AI validates:

  • operation_key uniqueness in existing registries
  • All entrypoints exist and have ABILITY.md with marker blocks
  • Implementation reference validity (if script, file exists or will be created)

Step 3: Preview Changes (AI)

AI runs orchestrator with --dry-run:

bash
python scripts/devops/scaffold/new_ability_low.py   --operation-key "<operation_key>"   --summary "<summary (10-200 chars)>"   --kind script   --impl-script "/scripts/my_ability.py"   --entrypoints "/modules/integration"   --dry-run --json

AI presents planned_changes to human.

Human checkpoint: Review and approve planned changes.

Step 4: Execute Registration (AI)

AI runs orchestrator without --dry-run:

bash
python scripts/devops/scaffold/new_ability_low.py   --operation-key "<operation_key>"   --summary "<summary (10-200 chars)>"   --kind script   --impl-script "/scripts/my_ability.py"   --entrypoints "/modules/integration"   --json

The orchestrator automatically triggers abilities.py --regenerate to update ABILITY.md indexes.

Step 5: Verify Results (AI)

The orchestrator automatically triggers index regeneration. AI verifies:

  • Registry entry exists in /.system/registry/low-level/
  • ABILITY.md indexes are updated
  • abilities.py --check passes

Step 6: Record in Workdocs (AI)

AI records scaffold execution:

  • Task ID, timestamp, parameters
  • Registry entry created
  • ABILITY.md files updated
  • Follow-up: implement the actual script/api/mcp if not yet done