Skip to content

AI_E2E_SCENARIOS.md

Latest commit

 

History

History
105 lines (82 loc) · 4.16 KB

AI_E2E_SCENARIOS.md

File metadata and controls

105 lines (82 loc) · 4.16 KB
Raw

中文 | English

AI End-to-End Execution Scenarios

This document provides executable AI storylines from routing to execution.

Scenario 1: Query aelf block height

Goal: user asks “check current chain block height”.

Execution flow:

  1. Read skills-catalog.json and locate chain-read skills.
  2. Use routing matrix and choose aelf-node-skill (instead of aelfscan).
  3. Run quick bootstrap:
./bootstrap.sh --only aelf-node-skill --skip-install
  1. Follow that skill repo's setup path (MCP/OpenClaw/CLI).
  2. Call the related read tool (for example get-chain-status / get-block-height).

Success criteria:

  1. Returns latest block height.
  2. Output includes selected skill and invocation mode for traceability.

Scenario 2: Transfer routing before execution (EOA/CA)

Goal: user asks “transfer 1 ELF” without identity mode.

Execution flow:

  1. Perform identity routing first:
  • default portkey-eoa-agent-skills when no CA signals exist.
  • switch to portkey-ca-agent-skills on guardian/CA hash/register/recover signals.
  1. Ensure active wallet context is set (portkey_set_active_wallet or create/unlock flow auto-write).
  2. Run read checks first (balance, address validity, network config).
  3. Execute write operation (transfer) with signerMode=auto.
  4. If context points to encrypted wallet, provide password input or password env.
  5. If user requests safety first, prefer simulate/dry-run when available.

Success criteria:

  1. Routing decision is explainable (why EOA vs CA).
  2. Required pre-write checks are done before sending tx.
  3. Output includes recovery hints (insufficient balance, network issues, missing context/password).

Scenario 3: Awaken swap composition (Wallet + DEX)

Goal: user asks “swap ELF to USDT on Awaken”.

Execution flow:

  1. Use routing matrix to determine wallet identity mode first:
  • no CA signal: portkey-eoa-agent-skills
  • with CA/guardian signals: portkey-ca-agent-skills
  1. Use wallet skill for pre-checks (balance, address, network) and confirm active context.
  2. Use awaken-agent-skills to fetch quote.
  3. Perform allowance/approve and swap with signerMode=auto (shared context first).
  4. If context signer requires password, pass tool input password or password env.
  5. Return tx outcome with explicit composition-routing explanation.

Success criteria:

  1. Clearly states the composition of wallet skill + awaken skill.
  2. Required pre-checks and permission checks are completed before swap.
  3. Output includes recovery hints (insufficient balance, approve failure, network errors).

Scenario 4: AI contributor onboarding a new skill

Goal: AI receives a request to integrate a new skill into this aggregation repository.

Execution flow:

  1. Read CONTRIBUTING.md and docs/AI_SKILL_CONTRACT.md.
  2. Validate required schemas:
  • docs/schemas/workspace.schema.json
  • docs/schemas/skill-frontmatter.schema.json
  • docs/schemas/openclaw.schema.json
  • docs/schemas/skills-catalog.schema.json
  1. If the skill supports MCP + setup, ensure catalog metadata exposes setupCommands.ironclaw, clientSupport.ironclaw, distributionSources, and clientInstall.
  2. Add new path to workspace.json with ${SKILLS_BASE} placeholder.
  3. If dependency exists, add direct dependsOn entries.
  4. Run gates in order:
bun run catalog:generate
bun run health:check
bun run readme:check
bun run security:audit
./bootstrap.sh --only <skill-id> --skip-install
  1. Prepare PR body with the 6 fixed sections from AI contract.

Success criteria:

  1. Catalog generation passes with schema 1.3.0.
  2. No gate failure (health/readme/security/bootstrap).
  3. PR description contains Goal/Non-goal, key files, contract mapping, validation outputs, risk, rollback.

Additional install-routing rule:

  1. GitHub repo URLs are discovery-only.
  2. For OpenClaw/IronClaw execution, prefer clientInstall.*.installCommand when present.

Common Recovery Template

  1. Dependency download failed
  • Action: ./bootstrap.sh --source github --only <skill-id>
  1. skill-id not found
  • Action: bun run catalog:generate, then retry --only
  1. health check failed
  • Action: follow health:check output and add missing artifacts (setup/mcp/openclaw)