Bo
7a765cde19
Executes docs/plans/2026-08-07-agentops-operations-layer-alignment.md: AgentOps is the operations layer for agentic engineering; the federated integration graph is the topology, the semantic work-and-proof protocol is the contract, and RPI is the standard one-experiment traversal. Retires the ao flywheel command family and all knowledge-flywheel product state, tombstones the seven-move operating-loop workflow, narrows ao init and the .agents state writers to declared destinations, renames the core architecture page to rpi-traversal.md with a compatibility redirect, aligns AGENTS.md, 25 skills, public and package copy, regenerates every owned projection, and strengthens the conformance gates with planted-negative proofs. Both the alignment subject and the follow-up gate-bookkeeping commit carry fresh author-distinct validation PASS verdicts with empty not_checked scope. Test-Removal-Reason: dead knowledge-flywheel and session-store surfaces were deleted with their tests (operations-layer alignment)
2.9 KiB
Agent workflow reference
This document expands the repository contract in AGENTS.md.
AgentOps is the operations layer for agentic engineering; the standard traversal through its federated integration graph is one pass (exact semantics: rpi-traversal.md):
RPI -> Plan -> Implement -> fresh Validate -> report and stop
1. Plan
Resolve one active behavior in the caller-owned tracker, issue, or conversation. Keep acceptance, important non-goals, required evidence, write scope, and the first useful check in that source. Do not create a second planning artifact.
The runtime snapshots the exact resolved source bytes under
.agents/ao/intents/sha256/<digest>.intent. This is derived identity, not a
model-authored packet, and makes conversation-only intent readable by a fresh
validator. The pure helper accepts a file or stdin:
python3 skills/validate/scripts/validate.py snapshot-intent --source PATH # use - for stdin
2. Implement
Run one bounded RED-GREEN-refactor experiment when the behavior supports it.
The runtime derives factual check receipts, actual changed paths, author context
ID, and subject-manifest.v1; the model does not transcribe them into a
candidate packet. Implement does not commit, claim, repair, retry, close, push,
or deliver.
3. Validate
An author-distinct context judges the exact subject against acceptance. It
returns PASS, FAIL, or NOT_PROVEN, lists checked and unchecked scope, and
stops. PASS requires nonempty checked scope, top-level evidence, and evidence
for every criterion. A NOT_PROVEN finding states the concrete missing runtime
precondition or examined uncertainty; it does not manufacture a next action.
Validate does not repair, re-plan, choose a next action, or authorize Git.
The returned result is sufficient for interactive use. Validate persists the
same result as content-addressed verdict.v2 only when the caller requests a
machine-readable artifact or a declared downstream consumer requires one.
4. Caller continuation
The caller receives the RPI report and decides what happens next. A revision updates the caller-owned intent source and starts a new invocation. RPI creates no parallel revision artifact. Changing acceptance changes the intent digest.
Optional surfaces
Premortem, postmortem, councils, idea genies, runtime adapters, research tools,
and factory dispatch are caller-selected. They never become hard dependencies
or lifecycle authorities. dispatch_once executes only explicitly supplied,
disjoint work once and performs no selection, retry, validation, integration,
Git, closure, or delivery.
Repository mechanics
Git branches, worktrees, trackers, pull requests, merge queues, CI, pushing,
rollback, and release are repository or caller policy. ao gate check may run
deterministic checks; it conveys no semantic verdict.