Integrate the discipline library into an actual AI coding project

What is already done, and what is not

The downloaded package contains role charters, standalone prompts, an advisory router and evidence-record checks. It does not install itself into a live repository, connect to Bubbles, read mail, edit DNS or configure an analytics account. No real project path was supplied or modified during this expansion.

Codex reads project instructions through its documented AGENTS.md hierarchy. Root instructions can be supplemented by nearer project instructions, subject to the documented discovery and size rules. Use an actual project root, not an unrelated downloads folder. Official Codex configuration.

Recommended layout

Place this extracted library inside an approved documentation location such as docs/seo-aio-library/, or keep it in a known read-accessible location and use a verified relative/absolute reference. The example is not a claim that this directory exists on your machines.

Merge the integration snippet into the project's existing AGENTS.md; preserve higher-priority instructions, tests and project safety rules. Do not overwrite a real project file blindly. Update the snippet's library path to the actual chosen location.

A repository-level instruction may say: read the library's operating model and route each task to a lead before making changes; use current source-linked modules over retained legacy advice; require evidence and explicit approval for risky changes. It should not say that every role has already verified the result.

Use a role context pack

From the extracted library root:

python3 Tools/discipline_router.py --task-file /path/to/task.txt --intent audit --output /tmp/route.json --prompt-output /tmp/task-context.txt

Replace the example path with a real authorized file. Review the proposed lead and consulted roles, then supply the context pack and task to the assistant operating on the actual project. The helper supports a lead override for incorrect or ambiguous keyword classification. It does not invoke Codex or any model, and it cannot automatically “prompt itself” without an agent reading and following the instructions.

The pack intentionally contains a small set of role prompts. Remaining consulted roles are named for follow-up, not claimed as executed. Let the agent read linked modules on demand rather than embedding the whole legacy corpus.

Connect actual evidence

Create a private project working directory for task intake, verified site facts, findings, raw artifacts, redacted artifacts and acceptance records. Keep secrets in the project's actual secret-management system, never in prompts or templates. Populate unknown facts from authorized inspection. Record the build/config version and evidence acquisition time.

When a task requires a real mailbox, actual Safari/iPhone, DNS control, production logs or analytics access, the agent must use available authorized tools or mark the check BLOCKED. A simulated environment or unit fixture cannot fill those gaps.

Validate a record

python3 Tools/evidence_gate.py --contract /path/to/approved-contract.json --record /path/to/acceptance-record.json --evidence-root /path/to/private-evidence

Use real paths and an approved contract. The template record is intentionally incomplete and should fail. Exit 0 means only that structural checks passed. A responsible owner must still verify evidence truth, actual independence and release/publication authority. Read acceptance rules.

Maintenance and compatibility

Keep the baseline 99-Originals/ as history. Do not execute old code examples because an AI found them. Preserve the current/legacy labels and the commercial tier definitions. Refresh fast-changing provider and reporting documentation before acting on it. Track local project overrides separately so future library updates cannot silently remove them.

Run the supplied tests and integrity auditor after changing the role registry or tools. A passing library test run proves the local tooling behavior tested, not production readiness of any website.