菜单
Execute and troubleshoot Supabase edge function `embedding_ft` that consumes PGMQ embedding jobs, calls AWS SageMaker embeddings, and writes vectors back to Postgres. Use when validating job payload handling, investigating failed embeddings, tuning ack semantics, or adjusting worker auth/environment.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/tiangong-lca/skills --skill embedding-ft命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | embedding-ft |
| description | Execute and troubleshoot Supabase edge function `embedding_ft` that consumes PGMQ embedding jobs, calls AWS SageMaker embeddings, and writes vectors back to Postgres. Use when validating job payload handling, investigating failed embeddings, tuning ack semantics, or adjusting worker auth/environment. |
TIANGONG_LCA_CLI_DIR or --cli-dir only for an exact local dev/CI override.tiangong-lca auth doctor-auth --json. If it returns login-required, stop and ask the human user to run tiangong-lca auth login in a trusted terminal.node scripts/run-embedding-ft.mjs with standard tiangong-lca admin embedding-run flags.tiangong-lca admin embedding-run; the CLI resolves and refreshes its private OAuth session.completedJobs and failedJobs, then triage via references.Never ask an AI user for a username, password, authorization code, access token, refresh token, or the deprecated encoded API key. Headless runs may use only an orchestrator-injected short-lived TIANGONG_LCA_ACCESS_TOKEN; do not pass it through argv, prompts, logs, or artifacts.
TIANGONG_LCA_API_BASE_URL="https://example.supabase.co/functions/v1" \
TIANGONG_LCA_SUPABASE_PUBLISHABLE_KEY="<publishable-key>" \
TIANGONG_LCA_OAUTH_CLIENT_ID="<public-oauth-client-id>" \
tiangong-lca auth doctor-auth --json
TIANGONG_LCA_API_BASE_URL="https://example.supabase.co/functions/v1" \
TIANGONG_LCA_SUPABASE_PUBLISHABLE_KEY="<publishable-key>" \
TIANGONG_LCA_OAUTH_CLIENT_ID="<public-oauth-client-id>" \
node scripts/run-embedding-ft.mjs --dry-run
TIANGONG_LCA_API_BASE_URL="https://example.supabase.co/functions/v1" \
TIANGONG_LCA_SUPABASE_PUBLISHABLE_KEY="<publishable-key>" \
TIANGONG_LCA_OAUTH_CLIENT_ID="<public-oauth-client-id>" \
node scripts/run-embedding-ft.mjs
node scripts/run-embedding-ft.mjs \
--input ./assets/example-jobs.json \
--base-url "https://example.supabase.co/functions/v1"
# Force a local CLI working tree during dev/CI
TIANGONG_LCA_CLI_DIR=/path/to/tiangong-lca-cli \
node scripts/run-embedding-ft.mjs \
--dry-run \
--base-url "https://example.supabase.co/functions/v1"
400: request body is not a valid job array.500: SageMaker request/response parsing failure.completedJobs < submitted: inspect queue payload, row version, and content function output.references/env.md: auth and caller environment.references/job-contract.md: queue semantics and DB side effects.references/testing.md: smoke-test and debug checklist.