Tools · Context

Context tools

The first tools any agent should reach for. They turn vague requests into ranked, source-labeled context — instead of grepping blind. All read-only.

context_packet read

Turn a messy coding request into ranked, source-labeled context using deterministic providers first.

When to use it

The first call for any vague task. "debug auth onboarding", "where does X live?", "review impact of changing Y" — start here.

Input

{
  "request": "debug why manager onboarding role checks are failing",
  "focusFiles": ["lib/auth/dal.ts"],   // optional
  "includeInstructions": true,
  "includeRisks": true,
  "includeLiveHints": true,
  "freshnessPolicy": "prefer_fresh",
  "budgetTokens": 4000
}

Output (abridged)

{
  "primaryContext":  [{ "path": "...", "reason": "...", "score": 0.91 }, ...],
  "relatedContext":  [...],
  "activeFindings":  [...],
  "risks":           [...],
  "scopedInstructions": [...],
  "recommendedHarnessPattern": "...",
  "expandableTools": ["route_trace", "db_rls", "reef_inspect"]
}

Related

reef_scout for broader fact/finding scout. reef_inspect for one-file deep dive. See What is a context packet? for the full shape.

reef_scout read

Reef-backed scout packet of facts, findings, rules, and diagnostic candidates for a request.

When to use it

When the question is "what does Mako already know about this?" rather than "what files are relevant?". Heavier on durable Reef data than context_packet.

Input

{
  "request": "audit auth boundaries before refactor",
  "limit": 30
}

Output

{
  "facts":           [...],
  "findings":        [...],
  "rules":           [...],
  "diagnosticCandidates": [...]
}

reef_inspect read

Explain one file or subject fingerprint with its facts, findings, and relevant diagnostic runs.

When to use it

Before editing a sensitive file. Returns the full evidence trail Reef has for that subject — what's known, what's questioned, what's been acked.

Input

{
  "subjectKind": "file",
  "subjectId":   "lib/auth/dal.ts"
}

evidence_confidence read

Label evidence as verified, fresh, stale, fuzzy, historical, contradicted, or unknown.

When to use it

When you've pulled a fact or finding and want to know whether you can trust it. Drives the freshness model — see How agentmako tracks freshness.

tool_batch read

Run independent read-only Mako lookups in one labeled request. Reduces MCP round trips.

When to use it

When you want freshness, conventions, and open loops in one call. tool_batch rejects mutation tools (no project_index_refresh, working_tree_overlay, diagnostic_refresh, db_reef_refresh, finding_ack, or finding_ack_batch).

Input

{
  "verbosity": "compact",
  "continueOnError": true,
  "ops": [
    { "label": "freshness", "tool": "project_index_status", "args": { "includeUnindexed": false } },
    { "label": "auth-conventions", "tool": "project_conventions", "args": { "limit": 20 } },
    { "label": "open-loops", "tool": "project_open_loops", "args": { "limit": 20 } }
  ]
}