Crate agent_decision_log 

agent-decision-log

WHY-layer decision log for AI agents. Records the reasoning behind each branch in an agent run: what options were considered, which one was chosen, why, and what happened afterward.

Sibling of agentsnap (CALLS) and agenttrace (COST + LATENCY). Together they cover the three audit dimensions of an agent run.

Quick example

use agent_decision_log::DecisionLog;
use serde_json::json;

let mut log = DecisionLog::new();
let id = log.add(
    vec!["search_web", "ask_user"],
    "search_web",
    "Query is specific enough to search without clarification.",
    json!({"turn": 3}),
);
log.set_outcome(&id, "Found 5 relevant docs.");

let d = log.find_by_id(&id).unwrap();
assert_eq!(d.chosen, "search_web");
assert_eq!(d.outcome.as_deref(), Some("Found 5 relevant docs."));

Round-trip to JSONL

use agent_decision_log::DecisionLog;
use serde_json::json;

let dir = tempfile::tempdir()?;
let path = dir.path().join("decisions.jsonl");

let mut log = DecisionLog::new();
log.add(
    vec!["a", "b"],
    "a",
    "a is cheaper",
    json!({}),
);
log.to_jsonl(&path)?;

let loaded = DecisionLog::from_jsonl(&path)?;
assert_eq!(loaded.len(), 1);

Structs

Decision

A single decision point in an agent run.

DecisionLog

Append-only log of agent decisions.