Agentic Craft

Observability

Activity timelines, token tracking, session history, and error logs for continuous agent oversight.

Timeline

Activity Timeline

A chronological stream of agent actions — tool invocations, messages, and state transitions — for full auditability of review workflows.

Threshold warningSource coverage is below release-review minimum

JSON export support is missing from the draft. The run stays in review until the owner resolves the gap or accepts the risk.

Controls

Live feed

Streaming

Checking Export workflow algorithm references

Just nowTool call

Cross-referencing Project brief v3 against Launch checklist

12s agoTool call

Identified 2 coverage gaps in Access rules requirements

34s agoMessage

Started requirement gap analysis for ACME Customer Portal

1m agoMessage

Run state

Agent	Status	Task	Progress	Confidence	Cost	Updated
Source Collector Retrieves documents	CompleteComplete	Found the source policy and project brief sections used in the answer.	x 100%	93%	$0.04	12s ago
Requirements Mapper Maps requirements to controls	WorkingWorking	Comparing the active request against source coverage.	x 72%	86%	$0.16	now
Document Drafter Drafts project brief sections	IdleIdle	Waiting for the next user or agent event.	x 0%	n/a	$0.00	now

Export review run trace4 events

Operational events, source touches, warning state, and queued handoff for the active review.

Collected launch sourcesRead the project brief, launch checklist, and triage policy.3 sources11:40 AM18s

The trace records touched source objects and completion state, not hidden reasoning.

Mapped export coverageCompared CSV-only export text against the current policy.Project brief v311:42 AM24s

Coverage threshold warningJSON export is missing from the draft, so the run stays in review.Launch Policy v211:43 AMvisible on hover

Warnings should include the affected object, source, and recovery path before escalation.

Queued reviewer handoffPrepared a packet for the launch owner to confirm ownership.Review sessionnow

Element	Details
Entry animation	Slide-in from below, 250ms ease-out, staggered 60ms per item
Live indicator	Pulsing dot with "Streaming" label when in live mode
Action types	Tool call and Message — quiet text after the timestamp; the row icon carries the type
Filtering	Filtered view shows only tool call entries for focused audit

The activity timeline is the audit surface for agent activity — every agent action is recorded with a timestamp, type classification, and human-readable description. Live mode streams entries as they occur; History and Filtered modes enable retrospective analysis during review team sessions.

Economics

Token Usage

A visual meter showing token consumption against the session or daily budget, with cost estimates and usage warnings.

Controls

Token budget

$0.86 estimated

12,400of 100,000 tokens

Well within budget12% used

Sessions

Avg. per session

4,133

Remaining

87,600

Element	Details
Usage bar	Height 12px, rounded-md, darkens above 80% threshold
Warning state	Appears above 80% with advisory text and subtle border highlight
Cost estimate	Dollar amount shown alongside token count — primary metric for non-technical users
Stats grid	3-column layout: sessions, average per session, remaining tokens

Token meters translate raw API consumption into a budget metaphor that product teams understand. The 80% warning threshold provides time to prioritize remaining review tasks before throttling begins — particularly important during time-bounded review team session sessions.

Conversation

Session Timeline

A vertical timeline showing the sequence of requests, tool calls, and agent responses within a review session.

Controls

Reviewer

Check if the project brief covers all Export workflow requirements for the ACME Customer Portal.

Agent

Searched document repository

I reviewed the Project brief v3 against Export workflow and found the export behavior specification references CSV-only export, but the latest Launch Policy v2 expects CSV and JSON exports. This is a gap that should be addressed before the review team session.

Element	Details
Connector	1px vertical line between nodes, bg-border color
Node icons	Arrow for reviewer, brain for agent — rounded-md 24px containers
Agent prose	Source Serif 4, 14px/22px, theme-aware foreground color
Tool annotations	Shown above agent response as muted text with search icon

Session timelines provide the conversation-level audit trail that review team reviewers need when reviewing how an agent arrived at its conclusions. Multi-turn sessions show how iterative refinement (e.g., narrowing scope from all requirements to just the integration and export sections) leads to more targeted analysis.

Diagnostics

Error Log

A filterable log of errors, warnings, and operational anomalies from the current review session with expandable detail views.

Controls

No errors in current session

All agent operations completed successfully since session start.

Element	Details
Empty state	Centered check icon with "No errors" message and subtitle
Error entries	Clickable rows with severity badge, timestamp, and expand/collapse
Severity levels	Warning and Error — shown as muted chip badges, no colors
Expanded detail	Animated expand with full error description and context

The error log provides the diagnostic transparency that activity review requires — reviewers can inspect every operational anomaly, understand its context, and verify that the agent's error handling meets the product's claimed readiness level.