search

name	search
description	Search the MIF research index — find findings by free-text query or by structured filters over the index fields (dimension, tags, namespace, verdict). Use this skill when the user wants to find research findings, look up what is known about a subject, query findings, or asks 'what do I have on X', 'find findings about Y', 'search research for Z', 'what research', 'what findings'. Reads the projected index, not raw tag files.
version	0.5.0
argument-hint	<query> [--lex|--sem] [--tag <tag>] [--namespace <ns>] [--dimension <dim>] [--verdict <v>] [--limit <N>]
allowed-tools	Read, Bash, Grep, Glob

search — query the MIF research index

Find findings across the corpus. Each entry in research-index.json is a flat projection of one MIF finding, carrying exactly:

{ "id": "urn:mif:...", "title": "...", "namespace": "...",
  "dimension": "...", "tags": ["..."], "verdict": "...", "citations": 1 }

The index is reproducible from the MIF findings at any time; it is not a tag-derived recomputation. Rebuild it with bash scripts/build-index.sh <findings-dir> if it is missing or stale.

Backends

• lexical (jq) — the default and only required backend. Filters and free-text matching run with jq over the index fields. Always available.
• semantic (optional) — the mif-rh engine's finding index (a SQLite embedding index, separate from research-index.json). When the mif-rh MCP server is available in the session (wired by the repo's .mcp.json; binaries installed by scripts/fetch-engine.sh), use its search tool for free-text queries and find_similar for by-meaning ranking, then look results up in research-index.json by id. The engine's finding index is built with bin/mif-rh-cli review --build-index (rebuilding research-index.json via scripts/build-index.sh does NOT build it); if the MCP tools are absent or reply index-not-built, fall back to lexical --lex and say so. Treat it as an optional accelerator, never a dependency.

Arguments

• Free text — words matched against title + tags (lexical by default)
• --lex — lexical only (the default; explicit form)
• --sem — request the optional semantic backend; fall back to --lex if absent
• --tag <tag> — findings carrying this tag
• --namespace <ns> — scope to one MIF namespace (one topic)
• --dimension <dim> — filter by a config-declared dimension
• --verdict <v> — filter by verification verdict (e.g. survived, weakened)
• --limit <N> — max results (default: 20)
• No arguments — show an index summary

Lexical search and filters

Build the select() from whichever flags are present, then sort and cap:

jq --arg q "QUERY" --arg tag "TAG" --arg ns "NS" \
   --arg dim "DIM" --arg v "VERDICT" '
  [ .findings[]
    | select($q == "" or ((.title + " " + (.tags | join(" "))) | ascii_downcase | contains($q | ascii_downcase)))
    | select($tag == "" or (.tags | index($tag)))
    | select($ns  == "" or .namespace == $ns)
    | select($dim == "" or .dimension == $dim)
    | select($v   == "" or .verdict == $v) ]
  | .[0:LIMIT]
' reports/_meta/sample-session/research-index.json

Pass "" for any flag the user did not supply. Point the path at the active topic's index, or at the sample (reports/_meta/sample-session/) when exploring.

Index summary (no arguments)

jq '{ count,
  by_dimension: [ .findings | group_by(.dimension)[] | { dimension: .[0].dimension, count: length } ],
  by_verdict:   [ .findings | group_by(.verdict)[]   | { verdict: .[0].verdict, count: length } ],
  namespaces:   [ .findings | map(.namespace) | unique[] ] }' \
  reports/_meta/sample-session/research-index.json

Output format

Present each finding as:

**<title>** (<namespace> / <dimension>)
Verdict: <verdict> | Citations: <citations>
Tags: <tags joined>
id: <id>

Group by namespace when results span multiple topics. End with "Showing N of M findings."

Handoffs

Need	Route to
Coverage gaps, clusters, stale findings	/discover
Relationships between findings/entities	/graph
Reason about or stress-test the evidence	/lab
Adversarial web verification of a claim	/falsify

Prerequisites

• research-index.json for the active corpus. If missing: bash scripts/build-index.sh <findings-dir>.

Examples

/search update propagation
/search --tag distribution --verdict survived
/search --dimension technical --namespace harness/example-topic
/search --sem "living template"
/search