Surgical Context

Surgical Context is SigMap's demand-driven retrieval mode. Instead of dumping whole signature blocks into your agent's upfront context, it points the agent at exact line ranges, lets it fetch bodies on demand, and emits only what changed between turns.

Three mechanisms work together:

Mechanism	What it does	Flag / tool
Line anchors	Every signature carries a path:start-end suffix	on by default (v6.11.0)
Two-tier index	Emit a tiny symbol-header index; fetch bodies on demand	--mode index
Delta context	Emit only signatures for files changed since a ref	--since <ref>

Line anchors

Since v6.11.0, extracted signatures carry the source location they came from:

def extract(filepath)  :268-338
function loadConfig(cwd) → object  :42-58

The :start-end suffix is a 1-based, inclusive line range. An agent that sees loadConfig :42-58 can read those 17 lines instead of re-opening the whole file — the core idea behind Surgical Context.

Coverage grows by release: TypeScript and Python top-level declarations first (v6.11.0); v6.13.0 adds JavaScript and per-member anchors — TypeScript/ JavaScript class methods and interface members now carry their own range (spanning the member body), so get_lines can target an individual method. Remaining languages are being added in subsequent phases.

--mode index — two-tier output

sigmap ask normally writes ranked signature blocks. With --mode index it emits only the symbol-header index — declaration heads plus their anchors, no parameter lists, no return types, no bodies:

sigmap ask "where is config loaded" --mode index

# SigMap Query Context (index mode)
> Symbol index only — fetch exact lines on demand via the `get_lines` MCP tool.

## src/config/loader.js
```
function loadConfig  :42-58
function detectAutoSrcDirs  :12-39
```

The agent reads this minimal map, then fetches the exact lines for the symbol it actually needs. Upfront context shrinks; nothing relevant is lost.

get_lines MCP tool

get_lines is the demand-driven workhorse. Given a file and a line range it returns the exact source — clamped to the file bounds, secret-scanned, and sandboxed to the project root:

{ "name": "get_lines", "arguments": { "file": "src/config/loader.js", "start": 42, "end": 58 } }

# src/config/loader.js:42-58
```
function loadConfig(cwd) {
  ...
}
```

It is one of SigMap's MCP tools (alongside read_context, query_context, get_impact, and others — see the MCP server guide).

Delta context — --since <ref>

--since restricts ask output to files changed since a git ref, so a steady-state turn carries near-zero context:

sigmap ask "finish the refactor" --since main
sigmap ask "what did I touch" --since HEAD~3 --mode index

Only ranked files that changed since the ref are emitted. Combine with --mode index for the leanest possible turn.

Budget-aware degradation

When the generated context would exceed maxTokens, SigMap now degrades gracefully: it collapses signature bodies to their anchors first (keeping the symbol and :start-end) before dropping whole files. The agent can still re-fetch any body via get_lines. The usual budget report still prints:

[sigmap] budget: collapsed bodies to anchors, reclaimed ~1840 tokens
[sigmap] budget: dropped 3 files to stay under 6000 tokens

Agent configuration

Point your agent's MCP client at the SigMap server, then let it call get_lines on the anchors it sees.

Claude Code / Cursor / Codex (mcp config):

{
  "mcpServers": {
    "sigmap": { "command": "npx", "args": ["sigmap", "mcp"] }
  }
}

Then generate anchored context and query in index mode:

sigmap                                   # generate context (anchors on)
sigmap ask "add a language extractor" --mode index

Measuring the savings

Numbers are never hand-typed — run the dashboard to see the Token Reduction panel for your own repos and the published benchmark:

sigmap --report      # opens the dashboard with the Token Reduction panel

See the Benchmarks overview for methodology.