Use SIFS from the shell when you need codebase context. The CLI is the reliable path; MCP tools are optional and should only be used when they are visible in the current agent session.

When to Use

• Use for local code search, symbol discovery, behavior tracing, related-code lookup, and indexed file/chunk inspection.
• Use before reading many files by hand when the task starts with "where is...", "how is...", "find the implementation...", or "what code handles...".
• Do not use for general web research, package documentation lookup, or non-code document search unless the user explicitly points SIFS at a source tree.

Start by discovering the local contract:

sifs agent-context --json

Search by intent, behavior, symbol, or exact text:

sifs search "authentication flow" --source <project>
sifs search "save_pretrained" --source <project> --mode bm25
sifs search "save model to disk" --source <project> --limit 10

Inspect indexed files and chunks before reading broad files:

sifs list-files --source <project> --limit 200 --json
sifs get src/auth.rs 42 --source <project>
sifs find-related src/auth.rs 42 --source <project>

Use --source <project> when the agent may not be running from the target checkout. Use --filter-path <repo-relative-path> for path narrowing and --limit for bounded results.

If MCP tools named search, get_chunk, or list_files are visible, they may be used for the same workflow. If they are missing, configured-but-invisible, or failing, fall back to the CLI immediately.

Bundled support files:

• references/commands.md contains command recipes.
• references/mcp.md explains MCP fallback rules.
• references/troubleshooting.md covers setup and stale-instruction checks.
• scripts/check-setup.sh verifies that sifs is on PATH and can print the agent contract.