> **Purpose.** This repository is agent-operated. This file is the single source of truth for how agents (e.g., Codex) must behave.
Sign in to like and favorite skills
Purpose. This repository is agent-operated. This file is the single source of truth for how agents (e.g., Codex) must behave.
The push/PR must be aborted if this script fails.bash scripts/agent_post_session_update.sh
On first contact in a repo: 1.Read VISION.md (top intent) and OVERVIEW.md (structure). 2.Execute:
bash scripts/agent_bootstrap.sh
This performs a deep scan (scripts/agent_scan.py), builds .agents/index.json, initialises .agents/priorities.json, seeds a modules/*/tasks.json for discovered modules, and validates all JSON against schemas.
1.Plan: Read .agents/index.json and .agents/priorities.json. Select next task using:
bash scripts/agent_next_step.sh
This returns the canonical {task_id, file} to execute next.
2.Execute: Implement the task fully (no placeholders). Update or create supporting docs as needed. 3.Write back context: •Append decisions and rationale to the relevant task’s notes[]. •If new cross-module knowledge emerges, add refs[] with exact file+line spans. •If new sub-work appears, split it into atomic tasks with clear acceptance criteria. 4.Housekeeping (MANDATORY):
bash scripts/agent_post_session_update.sh
This validates JSON, refreshes indices, updates ROADMAP snapshots, commits, and pushes. If it fails, fix and rerun. Do not open a PR until it passes.
•VISION.md — immutable high-level goals, guardrails, and anti-drift constraints. •OVERVIEW.md — current shape of the system, module map, and links to task files. •.agents/index.json — machine index of modules, docs, and link targets. •.agents/priorities.json — dynamic, ordered list of tasks (critical-path first). •.agents/modules/*/tasks.json — atomic, verifiable tasks per module. •scripts/ — the only allowed operational scripts. They must remain robust and idempotent.
•Deterministic acceptance tests (unit/integration) exist and pass. •Performance/complexity is appropriate for expected scale. •Security considerations addressed (inputs validated, secrets not hard-coded, least privilege). •Observability (logs/metrics) and failure handling implemented where relevant.
•At the end of each task, append an actionable next_notes[] with hints you would have wanted before starting the next task. •If priorities changed during your session, update .agents/priorities.json using the scripts. Never leave the repo in a state that requires implicit memory.
Compliance with this contract is required for any PR authored by an agent.