Artifact-bound delivery

Artifact-bound delivery means the work stays connected to the object that defines it. Every loop reads from an artifact and writes back durable, reviewable outputs tied to that same artifact.

What counts as an artifact

ClosedLoop.ai treats several object types as first-class:

• PRDs — the problem, outcome, constraints, and success criteria.
• Implementation plans — the task-level decomposition a loop executes against.
• Critic reviews — structured feedback on a plan from the code plugin's critics.
• Judge reports — structured grades on a plan, code, or PRD from the judges plugin.
• Code maps — a structural snapshot of the codebase the loop explored.
• Learnings — patterns, pitfalls, and conventions extracted from a run.
• Code review findings — multi-agent review outputs attached to a PR.

Each of these has a schema, a storage location, and a lifecycle.

Where they live

Session artifacts live in the session work directory ($CLOSEDLOOP_WORKDIR, typically .closedloop-ai/work/):

prd.md
plan.json
plan.md
plan-evaluation.json
investigation-log.md
requirements-extract.json
code-map.json
code-context.json
plan-context.json
judge-input.json
plan-judges.json
code-judges.json
prd-judges.json
state.json
log.md
perf.jsonl
visual-requirements.md
reviews/*.review.json
agents-snapshot/
.cross-repo-needs.json
.workspace-repos.json
.dev-environment.json
.learnings/...

Organization-level artifacts live under ~/.closedloop-ai/:

learnings/org-patterns.toon
learnings/closedloop-learnings.json
electron-port

Benefits

• Clearer review context — reviewers see the plan, the critics, the judges, and the diff together.
• Better handoffs — the next person (or agent) inherits the durable work, not a chat transcript.
• Reusable execution history — prior runs become training data for relevance scoring and pattern matching.
• Easier judgment against the goal — the judge reads the PRD, the plan, and the output, and scores per-metric.

Why this is different from chat

Chat workflows optimize for local velocity at the cost of organizational memory. Artifact-bound delivery optimizes for compounding leverage — every run leaves the organization better equipped for the next one.

That is a core difference between a delivery system and a chat workflow.