alphacouncil-agent AGENTS.md

Agent Instructions

Read CLAUDE.md first. It defines the workflow, analyst roles, final-report requirements, evidence contract and repository boundaries.

For report-generation behavior, also follow docs/report-contract.md. The chat handoff may be concise, but the saved report and artifact files must preserve the full evidence chain.

For anti-laziness governance, also follow skills/agent-skills-governance/SKILL.md. It is bundled with the plugin so installed agents inherit the same gates without separately installing addyosmani/agent-skills.

For methodology comparisons or Skill experiments, also follow skills/alphacouncil-method-lenses/SKILL.md. Its 26 on-demand references are provisional explanation material and never replace a frozen deterministic PersonaPack result.

Run npm run check after any code or prompt change.

Current Release Boundary

Package/plugin version 1.6.0 is the current source release candidate; GitHub, Codex marketplace and npm's public latest tag must be verified separately before claiming it matches this source. The current source tree carries 26 physical PersonaPack v3 packs, 26 operator_lens seats and 52 executable method tools, with 0 validated method_model seats. Human review of the authored formulas, the live four-code-host end-to-end run, and the separate ChatGPT Work developer-mode end-to-end run are outstanding, so do not present a passing full/quick report or a packaged smoke as evidence that any has been done.

Council Modes

full is the default and uses the full_v2 contract. It requires all eight mandatory evidence roles, every selected method, the three-round Bull/Bear cross-exam and PM. If any mandatory evidence role still fails after its one bounded parse-only repair, persist the failure/diagnostic artifacts, skip method/debate/PM model calls and terminate incomplete before downstream synthesis. Never auto-downgrade that run to quick.

Plugin-managed headless full runs launched with analyze_symbol use the selected pace's hard queue-to-terminal-persistence ceiling: 15 minutes for fast, 30 for normal, and 60 for slow. The eight mandatory evidence workers start in one parallel wave. Each selected v3 method first produces a deterministic, frozen stance and then gets one isolated voice worker that may explain, but never change, that stance. Bull and Bear run in parallel within each of the three rounds, with a hard barrier between rounds, followed by the PM. Retries, queueing and persistence consume the same deadline; callers and environment variables may lower it, never raise it. At expiry, persist a terminal fail-closed incomplete run and name every missing/skipped seat. The clock guarantees terminal persistence, not successful completion when data providers, search or model transport deteriorates.