fine-tuned-llm-tracing

Imported from williamntlam/fine-tuned-llm-tracing (AGENTS.md). Install upstream with npx skills add williamntlam/fine-tuned-llm-tracing. Copyright stays with the author.

Codex Instructions

Project

AI Incident Diagnosis Model Lab is a learning-focused LLM engineering project for training and evaluating incident-diagnosis models from synthetic traces, logs, metrics, topology, and incident metadata. Preserve deterministic, inspectable experiments and ground-truth labels.

Working conventions

• Read the relevant project documentation and any applicable local skill before changing code or data.
• Keep changes scoped to the user's request. Do not introduce a framework, external service, model, or dataset without a clear project need.
• Do not commit secrets, generated model weights, raw credentials, or large experiment artifacts. Prefer documented, reproducible generation steps.
• Validate the narrowest meaningful unit after a change and report the command and result.

Change history

Whenever a task makes a meaningful repository change, update the root-level CHANGELOG.md in that same change. Add a newest-first entry headed with the local Toronto date and time in the form YYYY-MM-DD HH:MM TZ — brief delivery name; refresh the timestamp immediately before the default-branch push when needed. Summarize the meaningful changes, affected areas, and verification; do not include secrets, generated artifacts, or a commit-by-commit transcript. Several changes delivered in one push may share one entry. Activate the local pre-push validator for a clone with bash scripts/install-git-hooks.sh; it blocks a default-branch push that does not include a CHANGELOG.md change.

Documentation and skill upkeep

For every task, decide whether the implementation changes a user-facing workflow, architecture, data contract, experiment procedure, or reusable agent workflow.

• Update the relevant documentation in the same change when it would otherwise become stale or when the new behavior needs to be discoverable.
• Update an existing skill, or add a focused new one under skills/, when the task establishes reusable, non-obvious guidance for future Codex work.
• Do not create documentation or skills solely to satisfy this check; record only durable information that materially helps a future contributor or agent.

Local skills

Project-specific skills live in skills/<skill-name>/SKILL.md. Use this index to select a skill, then read only the matching entrypoint and any resource it explicitly routes you to. Do not load every skill by default.

Each skill must keep valid YAML frontmatter with a concise, discriminating description. Add an index row whenever a new project skill is added.