Vector Recall

Who the node is. The id is the key the DOT pins.

id

vector-recall

name

Vector Recall

type

tool

phases

none declared

What it does, and the prose the agent is handed when the graph runs.

action31 words

Embed the strand's query and return the k nearest passages from the internal store, each carrying its document id and similarity score so the synthesizer can weigh how close it is.

spec105 words

Embed the query in the internal-corpus strand you are given on query, apply that strand's filters, and retrieve the twelve nearest passages from the vector store, discarding any whose cosine similarity falls below 0.7. Emit them on passages, nearest first, each carrying its stored text, its document id and its similarity score, so a reader downstream can weigh a close match against a distant one. Read only: never embed or write this run's own text back into the store. If fewer than twelve passages clear the threshold, emit the ones that do and stop, padding the budget with weaker matches is worse than returning less.in full above

model

whatever the graph supplies

agent

not named

skill

skills/vector-recall.md

tools

Vector store

mcp

qdrant

params

k: 12, similarity_threshold: 0.7

What arrives, what leaves, which nodes it expects to hear from, and what may not.

inputs

query : plan

outputs

passages : json

dependencies

retrieval-planner

cannot

no type is refused

will_not

write this run's own text back into the store

The keys the static analysis reads. Nothing here instructs the agent.

risk_markers

none

notes32 words

Read-only against the store: the desk recalls what has already been indexed and never writes a run's own output back, so one bad answer cannot poison the corpus for the next question.

The card's own version, and who wrote it.

version

1.0.0

author

autogen

provenance

not stated