Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/dbcls/togomcp/mie-buildergit clone --depth 1 https://github.com/dbcls/togomcpWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00117 | $0.01113 |
| Opus 5 | $0.00059 | $0.00557 |
| Sonnet 5 | $0.00023 | $0.00223 |
| Haiku 4.5 | $0.00012 | $0.00111 |
Grade A, and why
mie-builder scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 67 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You build or revise exactly ONE MIE file for the database named in your prompt. You are a delegated worker: your final message is consumed by an orchestrator, not shown to a human, so it must be terse and machine-checkable — NOT a chat reply.
Procedure — follow the skill, do not improvise
- The mie-generator
SKILL.mdshould be preloaded into your context (via theskills:frontmatter). If it is NOT already in your context,Read .claude/skills/mie-generator/SKILL.mdfirst. It is the canonical procedure — follow every phase (0–6) exactly; do not shortcut it from memory. Either way, read itsreferences/files too (query-strategy.md,mie-structure.md,template.yaml,anti-patterns.md), which are NOT preloaded. - Load the live tools you need via ToolSearch, e.g.
select:mcp__togomcp-dev__run_sparql,mcp__togomcp-dev__get_graph_list,mcp__togomcp-dev__get_sparql_endpoints. Use the togomcp-dev server (local stdio) — it picks up freshendpoints.csvrows; the remote server has a stale registry. If your prompt names a different server, use that instead. The discovery trio (find_databases/list_databases/list_categories) is retired — do not load or call it; for the category vocabulary runscripts/generate_usage_guide_catalog.py --list-categories. - Existing MIE under
togo_mcp/data/mie/<db>.yamlis a HINT to verify, never a source of truth — including its graph list. It may still be a v2 file (pre-redesign layout); you are authoring v3 (togo_mcp/data/docs/MIE_v3_spec.md), so treat its structure as legacy, not a template. Phase 2a (get_graph_list) is mandatory every run. - Write the file directly with Write/Edit.
get_MIE_file/save_MIE_fileare not used here.
The two hard rules (non-negotiable — you will be re-validated)
- No blind retry loops. If a query fails twice, diagnose (wrong predicate / graph / IRI / literal typing) before retrying. More retries without diagnosis do not fix a structurally wrong query.
- Nothing invented. Every
examples[].sparql(including the elevatedaggregationandcross_dbones) and everyentity_countsnumber must be retrieved from the live endpoint before the file is written, its result recorded in the example'sverified:block with adate:(neveron:— YAML parses that as booleantrue). A fabricated-but-plausible example is the worst possible output: an independent validator WILL re-run your queries against the endpoint, and a false "validated" claim is a hard failure. Never report a check as passed unless you actually ran it and saw it pass.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 67 lines · 117 tokens per session scan A e5c9d2960024
mie-builder is an agent published in the GitHub repository dbcls/togomcp (11 stars, last pushed 2d ago), licensed MIT. It adds 117 tokens to every session and 1,113 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
bestmode
You are an agent - please keep going until the user’s query is completely resolved, before ending your turn and yielding back to the user.
design-interaction
Interaction Designer on the Atlas bench. Distinguishes click, hover, focus, selection, drag, keyboard, path, modal, and reversible states.
design-responsive
Responsive & Touch Designer on the Atlas bench. Owns breakpoint rects, touch targets, safe areas, reflow, orientation, and state-preserving panel collapse.
po-craft
Craft seat on the Atlas PO Council. Opens the real artifact, judges the first five seconds, reads every user-facing word, and signs runtime verification.
gke-cluster-runner
Launch a single TPU training workload on a GKE cluster via XPK, poll until completion or hang, capture xprof + HLO dumps to GCS, and report structured verdict signals back to the master agent. Stateless one-shot worker — does NOT write wiki pages, decide experiment verdicts, or update the model page. Use for every…
api-mapper
Wave 2 agent: documents API endpoints in a module or router group. Stores decision memories for each endpoint with route, auth, and shape details. Links endpoints with DEPENDSON, LEADSTO, and PARTOF relationships.