| name | simplicio-runtime |
| description | Tuple-Space + Yool safe-speed runtime for coding tasks. Use when the user asks to implement, refactor, fix, build, or explain code in a repository that vendors the simplicio-prompt spec (presence of YOOL_TUPLE_HAMT.md, kernel/yool_tuple_kernel.py, or a simplicio-prompt block in CLAUDE.md / AGENTS.md). Routes any user input through a Hilbert-indexed tuple graph with lazy batch_spawn, LaneWorkerPool, receipt cache, jittered backoff, circuit breakers and context compression. Scales to 1,000,000+ subagents without enumeration. |
simplicio-runtime
Apply the Tuple-Space + Yool execution contract to the active task.
When to use
Trigger when ANY of the following is true:
- The working tree contains
YOOL_TUPLE_HAMT.md or
kernel/yool_tuple_kernel.py.
CLAUDE.md, AGENTS.md, .cursorrules, or
.github/copilot-instructions.md contains the marker
<!-- simplicio-prompt:start -->.
- The user explicitly invokes
/simplicio or asks for the runtime by name.
Do NOT trigger when the user explicitly stands down with phrases like "stop",
"cancel", "exit runtime", or "ignore the simplicio-prompt".
What to read first
YOOL_TUPLE_HAMT.md
kernel/yool_tuple_kernel.py
prompts/agent-runtime-execution-prompt.md
guardrails/cpu_throttle.py, guardrails/disk_gc.py
examples/python/minimal_bus.py, examples/python/receipts.py
scripts/build_hamt.py
If any are missing, fetch the canonical spec from
npx simplicio-prompt --raw or
https://github.com/wesleysimplicio/simplicio-prompt.
Execution contract
For every user input X (no trigger keyword needed):
- Decompose X into an explicit Hilbert-indexed tuple graph.
- Create a root tuple at level 0.
- Use
batch_spawn(depth, branching, compression_threshold) for hierarchical
fan-out. depth=4, branching=32 ⇒ 1,048,576 subagents materialized only
when visited. Never enumerate them in output.
- Use
spawn_agent only for active materialized work.
- Route everything through
out_tuple, in_tuple, rd_tuple,
route_packet, scan_index.
- Apply
hookwall(wall_id, capability, action) before privileged work.
- Apply
compress_token and prune_idle to inactive subagents.
- Use
LaneWorkerPool and respect:
YOOL_TUPLE_LANE_CONCURRENCY=32
YOOL_TUPLE_MAX_LANE_CONCURRENCY=64
YOOL_TUPLE_CPU_QUOTA_PCT=95 (never raise above 100)
YOOL_TUPLE_QUEUE_MAXSIZE=8192
YOOL_TUPLE_COMPRESSION_THRESHOLD=1024
- Safe-speed path before any provider/LLM call:
- receipt/input-hash cache
- adaptive lane concurrency
- jittered backoff
- provider circuit breakers
- small-task batching
- prompt/context compression
- local yool routing for deterministic work
- speculative execution only for
idempotent=True tuples
- Honor safe-speed env vars when present:
YOOL_TUPLE_CACHE_MAX_ENTRIES, YOOL_TUPLE_CACHE_TTL_S,
YOOL_TUPLE_API_MAX_RETRIES, YOOL_TUPLE_API_BACKOFF_BASE_MS,
YOOL_TUPLE_API_BACKOFF_MAX_MS, YOOL_TUPLE_CIRCUIT_FAILURE_THRESHOLD,
YOOL_TUPLE_CIRCUIT_COOLDOWN_S, YOOL_TUPLE_BATCH_SMALL_TASK_SIZE,
YOOL_TUPLE_CONTEXT_COMPRESSION_CHARS.
Output
Status output is opt-in. Default: silent (final result only).
Enable with YOOL_TUPLE_STATUS=true. When enabled, return exactly:
[Tuple Space Snapshot]
[Active Agents/Subagents]
[Total Agents/Subagents]
[Proximo Yool a executar]
[Resultado parcial]
Per-field overrides (each default false):
YOOL_TUPLE_STATUS_SNAPSHOT, YOOL_TUPLE_STATUS_ACTIVE,
YOOL_TUPLE_STATUS_TOTAL, YOOL_TUPLE_STATUS_NEXT,
YOOL_TUPLE_STATUS_PARTIAL.