Agent sprint
Design dependable coding agents with guardrails and telemetry before launch.
0 of 4 checkpoints complete • 0%
Complete the focus area plan
Track deliverables, evaluations, and storytelling artefacts in one place.
Action checklist
PRD and guardrail brief submitted
Use AI CoE template plus NIST mapping.
Agent playbooks committed
Builder, Critic, Coach, Archivist, Companion YAML files.
Evaluation harness configured
Run scripts/eval-harness with realtime, perception, and preparedness suites.
Field test journal updated
Capture learnings, mitigations, telemetry references.
Success signals
- Guardrail detection rate >= 90%
- Telemetry traces exported to Supabase workspace
- Sponsor summary delivered with evidence links
Keep momentum across delivery and storytelling
Launch evaluations, prep sponsor updates, or refresh assistant prompts without leaving the workspace.
Run evaluation harness
Launch scripts/eval-harness for realtime, perception, and preparedness suites to validate the latest build.
Prep sponsor update
Assemble progress digest, telemetry screenshots, and decision log for executives.
Sync assistant prompts
Refresh Scout, Coach, Critic, Archivist, and Companion prompt packs with the latest artefacts.
Equip Scout, Coach, Critic, Archivist, and Companion
Copy-ready instructions keep agents aligned with the learner workspace and Supabase data.
Scout briefing
Summarise changes to modules AE-101, RP-203, and OR-502. Include risks flagged by OpenAI, Anthropic, or DeepMind this week.
Coach session
Review the agent playbook YAML and propose improvements to tool selection, success metrics, and mitigation prompts.
Critic eval run
Execute the latest evaluation config with realtime freshness and perception checks. Highlight regressions versus baseline metrics.
Archivist export
Generate Evidence Locker summary for the current focus. Include PRD link, evaluation run IDs, telemetry references, and CTA for stakeholders.
Where artefacts live
Link deliverables so assistants, mentors, and sponsors can verify progress.
PRD and guardrails
docs/evidence/prd/
Mission brief, guardrail assumptions, and acceptance criteria for the active sprint.
Evaluation runs
scripts/eval-harness/runs/
Toxicity, hallucination, realtime freshness, perception, and preparedness metrics exported as JSON and markdown.
Sponsor digest
docs/evidence/digests/
Narratives, outcomes, ROI signals, and next steps for executives and patrons.
Data sources powering the workspace
Supabase keeps state, Langfuse captures traces, and Evidence Locker stores artefacts.
Supabase learner hub
Stores module progress, evaluation runs, sponsor digests, and creator assets with row level security.
Langfuse and OpenTelemetry
Captures assistant traces, guardrail verdicts, and latency SLAs for realtime dashboards.
Evidence Locker
Versioned artefact repository linked to curriculum modules, playbooks, and compliance reports.
What we are wiring next
Help us stand up Supabase auth, telemetry widgets, and assistant sync.
Supabase onboarding
Provision Supabase project with auth, row level security, and tables for learners, progress checkpoints, evaluation runs, and sponsor briefs.
Workspace telemetry widgets
Embed Langfuse dashboards and Supabase chart components directly in the workspace once APIs are wired.
Assistant live sync
Enable real-time updates so Scout and Archivist reference Supabase data without manual refresh.