Claude Loop Goal

Claude Loop Goal (operating mode)

Claude-owned. The self-paced standing goal Claude runs each pass. Set by Ray 2026-06-02 after Phase 1 (self-select the emptiest ledger cell) reached completion — every subsystem's Map / Perf / JIP-HC / Drift dimension is source-reviewed (DR-1..DR-40); the only residual 🟡 cells are Auth/PV owner decisions, each backed by a source-cited finding + fix in Deep-review findings. This page records the operating mode; the role is in Claude long-term goal, the scoreboard in Codebase coverage ledger.

Mode: collaboration-follow, with research autonomy, self-paced

Work under Codex's lead. Codex is the integrator and navigator; each pass, read the shared coordination state and follow Codex's lead rather than self-selecting independent work — but you retain research autonomy to pull your own threads where they add value.

Per pass

1. Sync + read. The live GitHub wiki is the source of truth (clone C:\Users\Steff\_wasp_wiki_tmp; in-repo mirror branch docs/developer-wiki-claude in C:\Users\Steff\a2waspwarfare-docs). Pull-rebase, then read agent-events.jsonl, agent-collaboration.json, Coordination-Board.md, Progress-Dashboard.md, Agent-Worklog.md, agent-context.json, Codebase-Coverage-Ledger.md, Deep-Review-Findings.md, and any atlas/page Codex changed recently.
2. Integrity check. If a finding/cell you previously landed has regressed (e.g. a commit dropped by a concurrent force-push), recover it — git merge-base --is-ancestor <sha> HEAD to detect, git cherry-pick from the object store to restore; the mirror branch is the durable backup.
3. Follow Codex's lead. Pick up whatever Codex has handed off, requested, flagged as a candidate, or left open in those docs. Always verify against the Chernarus source mission (Missions\[55-2hc]warfarev2_073v48co.chernarus) before making any claim. Adversarially confirm/refute Codex's candidates at source; produce the source-verified finding that backs whatever page Codex is building.
4. Research autonomy. You are not limited to Codex's queue. Between or alongside following Codex's lead — especially on idle passes, or whenever you spot a thread worth pulling — you may self-direct: adversarial re-verification of existing findings, cross-cutting source archaeology into under-reviewed corners, tracing a hazard class end-to-end, cross-checking behavior against Bohemia Interactive Arma 2 OA 1.64 scripting docs (never Arma 3), or designing concrete remediations for the Auth/PV owner-decision items. Codex's explicit handoffs take priority; autonomy fills the gaps and follows your own leads.
5. Publish (only after review gates pass — JSON/JSONL valid, links resolve): append claim/finding/complete events, update the ledger cell + Deep-Review-Findings.md + Agent-Worklog.md + agent-context.json, commit a small claude:-prefixed wiki commit, pull-rebase (keep BOTH sides on append-only conflicts), push wiki master + sync the docs/wiki mirror.

Lane & constraints

• Stay in your lane: Codex owns navigation, mirror parity and atlas pages — leave integration-ready handoffs, don't rewrite them.
• Docs-only: no edits to the mission source unless Ray rescopes the loop to write remediation patches.
• Arma 2 OA 1.64 reference docs only, never Arma 3.

Done

Nothing is pending from Codex and every open ledger item is an explicit, owner-only decision (each already linked to a source-cited finding with a concrete fix).

Continue Reading

Previous: Claude long-term goal | Next: Codebase coverage ledger

Main map: Home | Findings: Deep-review findings | Agent file: agent-context.json