rag-cli Operator

Purpose

Run this project's local RAG workflow safely and consistently using uv run rag-cli ....

When To Use

Use this skill when requests mention:

• rag-cli
• uv
• init, ingest, query, or docs
• document indexing/search in this repo
• tool-friendly JSON outputs (--format json)

Command Order

Follow this order unless the user asks otherwise:

1. uv sync
2. uv run rag-cli init --workspace <path>
3. uv run rag-cli ingest --workspace <path> --src <path> [--recursive] [--ext ...] --format json
4. uv run rag-cli query --workspace <path> --question "<question>" --mode hybrid --top-k 3 --format json
5. uv run rag-cli docs --workspace <path> --format json

Standard Command Templates

uv sync
uv run rag-cli --help
uv run rag-cli init --workspace .
uv run rag-cli ingest --workspace . --src ./docs --recursive --ext "pdf,docx,md" --format json
uv run rag-cli query --workspace . --question "What is SPI timing?" --mode hybrid --top-k 3 --format json
uv run rag-cli docs --workspace . --format json

Output Contract

Default to --format json when output will be consumed by an agent/tool.

Expected success shape:

• answer: string
• snippets: [{ doc, section, text }]
• meta: object

Expected error shape:

• error: { code, message, details }

If command output is for humans only, --format text is acceptable.

Workspace And Config Rules

• Always pass --workspace explicitly when not operating in repo root.
• init creates .rag/ and baseline config/index state.
• Required environment/config keys:
  • POE_API_KEY (preferred)
  • OPENAI_BASE_URL (default: https://api.poe.com/v1)
  • LLM_MODEL (default: gemini-3.1-flash-lite)
  • EMBEDDING_MODEL (optional)
• Backward compatibility:
  • OPENAI_API_KEY is accepted if POE_API_KEY is not set.

PowerShell Machine-Level Env Setup

[Environment]::SetEnvironmentVariable("POE_API_KEY", "your-key", "Machine")
[Environment]::SetEnvironmentVariable("OPENAI_BASE_URL", "https://api.poe.com/v1", "Machine")
[Environment]::SetEnvironmentVariable("LLM_MODEL", "gemini-3.1-flash-lite", "Machine")
# Optional if your provider supports embeddings:
[Environment]::SetEnvironmentVariable("EMBEDDING_MODEL", "", "Machine")

Agent Integration Notes

Keep parameter naming aligned with:

• skills/search_docs_cli.json
  • workspace, question, mode, top_k
• skills/index_files_cli.json
  • workspace, path, recursive, ext

When mapping tool calls, preserve these names to avoid wrapper drift.

Troubleshooting

Missing .rag index

Symptoms:

• docs or query fails with missing index messages.

Action:

1. Run uv run rag-cli init --workspace <path>
2. Re-run ingest
3. Retry query or docs

Empty ingest results

Symptoms:

• ingest reports no files found or indexed count is zero.

Action:

1. Check --src path exists.
2. Confirm extension filter via --ext.
3. Add --recursive for nested folders.
4. Re-run ingest with --format json and inspect meta.

Missing API/env values

Symptoms:

• model/provider errors when running query pipelines that require remote backends.

Action:

1. Set required env keys.
2. Re-run from same shell/session.
3. Verify with a simple query call.

Non-zero exit code

Action sequence:

1. Re-run same command with --format json.
2. Parse error.code and error.message.
3. Apply minimal fix.
4. Re-run only the failed step, then continue command order.