ask

sigmap ask is the daily command in v5.5. It turns a natural-language coding question into a focused mini-context plus useful metadata.

sigmap ask "fix the login bug"
sigmap ask "explain the auth flow" --json

What it does

ask combines several steps that used to be separate:

1. Detects the query intent (debug, explain, review, refactor, search)
2. Ranks files against the current signature index
3. Writes a focused context file to .context/query-context.md
4. Prints the ranked files and one summary line — coverage, risk, and estimated cost are one flag away (--verbose, or --json for scripts)

Typical output

1  src/security/redact.js    14.289  131 tok
2  src/cache/entry.js        13.861  255 tok
3  src/security/scanner.js   12.916   39 tok
4  packages/core/index.js     9.235  220 tok
5  src/security/patterns.js   7.123   41 tok
5 of 529 files · cutoff 7.123 · 744 tokens · sha256:7e1624d83f1c → .context/query-context.md

Each row is a rank, a path, a score and the tokens that file's signatures cost in the index. The signatures themselves — with their :start-end line anchors — are in .context/query-context.md. The last line is what makes the result auditable: how many files were selected, where the cut fell, the size of the context and a hash of it. A second line appears only when something needs your attention, for example a file in the selection that has uncommitted changes.

For the full header — intent, coverage, risk and cost — add --verbose, which prints the same rows and then the box ask printed before v8.78.0 (the summary line is not repeated):

1  src/security/redact.js    14.289  131 tok
2  src/cache/entry.js        13.861  255 tok
3  src/security/scanner.js   12.916   39 tok
4  packages/core/index.js     9.235  220 tok
5  src/security/patterns.js   7.123   41 tok
────────────────────────────────────────────
 sigmap ask  "how are secrets redacted"
 Intent    : search
 Context   : 744 tokens  →  .context/query-context.md
 Selected  : 5 of 529 file(s) (--top 5) · cutoff score 7.123
 Hash      : sha256:7e1624d83f1c
 Selection : 5 source
 Coverage  : readable 100% (205/205 files in srcDirs)
 Risk      : NONE (0 file(s) changed vs HEAD)
 Cost      : $0.0019/query  (was $0.0133 · saved 86%)
           est. @ gpt-4o $2.5/Mtok input as of 2026-10-04; cost tokens est. (chars/4); "was" = full ranked files
────────────────────────────────────────────

Best prompts to use

• "explain the auth flow"
• "where is the middleware stack configured"
• "fix the race condition in UserService.findById"
• "review the retry logic in the queue worker"

The more task-shaped the query is, the better the ranking signal becomes.

JSON mode

Use --json when you want to feed the result into scripts, dashboards, or CI:

sigmap ask "explain the ranker" --json

The response includes the detected intent, ranked files, coverage, cost, and risk fields.

Recommended follow-up

After ask, use:

• validate when you want to sanity-check coverage
• judge when you want to score whether an answer was actually grounded in that context
• learning when you want to reinforce or penalize the files that helped