aziel-runtime

Adaptive recall

memory_recall

Read-onlyIdempotent

Instructions

Ranked adaptive recall after ChainLock verify (AKM-TRIAD-1.0). Belief list — not raw grounded stamps, not a public corpus cite, not one-id explain. Use this when you want a ranked belief list after verify, not raw grounded stamps. Do not use it for grounded ChainLock recall, library search, or explaining one memory_id; use chainlock_recall, library_lookup, or memory_get instead. Does not authorize action (authorizes_action=false). Do not treat posterior rank as fact (belief_is_not_truth). Failed ChainLock verify refuses CHAIN_VERIFY_FAIL and does not invent cards. Empty grounded recall bubbles refuse=no-stamp. Ranking is capped at 16 cards. Omit depth to rank at 5 (full budget), unlike chainlock_recall which defaults to 1. q/query is lexical rank text, not a SQL filter. Empty q still verify-then-ranks stored cards. use_case weights triad_fit; it is not a permission. Optional limit clips the already-capped list. Returns ranked cards after verify (count, facts, belief_is_not_truth, authorizes_action=false).

Input Schema

TableJSON Schema

Name	Required	Description
`q`	No	Optional lexical rank query (subject/fact). Alias: query. Empty still runs verify-then-rank; it does not invent facts.
`depth`	No	Optional ChainLock recall depth after verify. Omit for 5 (full budget). 0 is tip-only ranking.
`limit`	No	Optional result cap. Hard ceiling is 16 (MEMORY_CONTEXT_CAP) even if a larger number is sent.
`use_case`	No	Optional use-case label that weights triad_fit in ranking. Not a permission and not a truth claim.

Output Schema

TableJSON Schema

Name	Required	Description
`code`	No	FragGate or fabric code when present: FG-OK, FG-HALLUC-TOOL, FG-STUB, FG-LOCAL-ONLY, FG-UNKNOWN-OP, FG-GATE-REFUSE, FG-LAMB-REFUSE, or a module refuse such as MESH-* / AKM-*.
`door`	No	Door name. The public door is fraggate.
`ran_in`	No	Execution locale (for example aziel-runtime) when present.
`result`	No	Adaptive recall body: ok, adaptive=true, verified, count, facts (ranked cards with score/retrieval), belief_is_not_truth, authorizes_action=false. Refuses CHAIN_VERIFY_FAIL or no-stamp.
`status`	No	HTTP-like status when present on wrappers (200 ok; 400+ error / refuse).
`display`	No	Human-facing envelope. Show title and summary, then take the next input.
`receipt`	No	Optional receipt, ledger tip, or TemporalLock/ForgeReceipts exit when the door stamped one.
`refusal`	No	Explicit refuse object, code, or message when the door or engine refused.
`engine_op`	No	Resolved engine op when present (often inside result).
`ledger_tip`	No	Ask/refuse ledger tip when the door stamped one.
`provenance`	No	Provenance / input packet when the pipeline attached one.
`session_id`	No	Raw session id when session plumbing was used. Hidden unless the user asked for the chain.
`engine_slug`	No	Resolved engine slug when present (often inside result).
`limitations`	No	Capability limitations or Remain-OFF notes when present.
`engine_digest`	No	64-hex engine_digest when a true in-process engine ran (often inside result).

Schema Changelog

Changes observed during successful MCP inspections.

Changed6 schema fields changedv2.0.1Sep 14, 2026

changedInput schema / description

Previous value: -"All fields optional. depth follows ChainLock 0–5."New value: +"All fields optional. Default depth is 5. Empty q still runs verify-then-rank and does not invent facts."

changedInput schema / properties / depth / description

Previous value: -"Optional recall depth 0–5 after ChainLock verify."New value: +"Optional ChainLock recall depth after verify. Omit for 5 (full budget). 0 is tip-only ranking."

addedInput schema / properties / limit

Added value: +{
+  "description": "Optional result cap. Hard ceiling is 16 (MEMORY_CONTEXT_CAP) even if a larger number is sent.",
+  "maximum": 16,
+  "minimum": 1,
+  "type": "number"
+}

changedInput schema / properties / q / description

Previous value: -"Optional query to rank against. Empty query still runs verify-then-rank; it does not invent facts."New value: +"Optional lexical rank query (subject/fact). Alias: query. Empty still runs verify-then-rank; it does not invent facts."

changedInput schema / properties / use_case / description

Previous value: -"Optional use-case label that weights ranking. Not a permission."New value: +"Optional use-case label that weights triad_fit in ranking. Not a permission and not a truth claim."

changedOutput schema / properties / result / description

Previous value: -"FragGate body: ok, code, door, slug, op, engine_digest, ran_in, provenance, refusal, limitations, receipt, plus the engine result. Unknown names refuse FG-HALLUC-TOOL; stubs refuse FG-STUB."New value: +"Adaptive recall body: ok, adaptive=true, verified, count, facts (ranked cards with score/retrieval), belief_is_not_truth, authorizes_action=false. Refuses CHAIN_VERIFY_FAIL or no-stamp."

Addedv1.6.2Sep 12, 2026

TDQS

A4.3/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare read-only/idempotent/non-destructive, but the description adds substantial non-obvious behavior: it does not authorize action, posterior rank is not truth, failed verify refuses CHAIN_VERIFY_FAIL without inventing cards, empty grounded recall returns no-stamp, and ranking is capped at 16 cards. These are exactly the failure-mode and safety traits an agent needs.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness3/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The routing and constraints are front-loaded, but the text is repetitive — "Belief list — not raw grounded stamps..." is restated almost verbatim in the next sentence. Several sentences could be merged without loss, and the jargon-dense framing slows scanning.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a 4-param tool with full schema coverage and an output schema, the description covers behavior, routing, failure modes, and defaults thoroughly. It arguably over-explains the return shape (count, facts, belief_is_not_truth) that the output schema already provides, so nothing essential is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so the baseline is 3, but the description adds real meaning: q is lexical rank text (not a SQL filter) and empty q still verify-then-ranks, depth omitted defaults to 5 unlike chainlock_recall's default of 1, use_case weights triad_fit and is not a permission, and limit clips the already-capped list. This goes beyond the schema text.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose4/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description names a specific operation (ranked adaptive recall of a belief list after ChainLock verify) and explicitly distinguishes itself from siblings chainlock_recall, library_lookup, and memory_get. The core purpose is identifiable, though heavy jargon ("AKM-TRIAD-1.0", "grounded stamps") adds noise without adding precision.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicit when-to-use ("Use this when you want a ranked belief list after verify") and explicit when-not with named alternatives ("Do not use it for grounded ChainLock recall, library search, or explaining one memory_id; use chainlock_recall, library_lookup, or memory_get instead"). Routing is unambiguous.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.