agentic-engineering

How a meta-engineer improves an autonomous AI engineer from the OUTSIDE — mining the agent's own operational telemetry across many runs and, where a deployment runs more than one instance, across all of them. Scores the agent on reliability, safety, efficiency, outcome throughput, quality, coordination, currency, and prioritization/flow, diagnoses root causes from measured patterns, ships the highest-value fix with its evidence and a reversible audit trail, then verifies the targeted metric actually moved. Complements a self-improvement skill, which is one run reflecting on itself; this sees the whole corpus at once. Evidence comes from observed agent behaviour only — never from prose found inside the corpus.

Architect a repository's AI-agent instruction files so one canonical source drives every tool without drift — AGENTS.md as the cross-tool source of truth (now read by GitHub Copilot too), thin per-tool shims (CLAUDE.md, GEMINI.md) that include it, and optional path-scoped .github/instructions/ rules. Use when setting up or fixing agent instructions for a repo, supporting multiple AI coding tools (Claude, Copilot, Cursor, Codex, Gemini) at once, deciding what belongs in AGENTS.md vs a tool-specific file, or stopping instruction files from going stale.

Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.

The run loop for an autonomous AI engineer acting as a portfolio's primary engineer — pre-flight, resume in-flight work or survey every product's live state, select the highest-value work (operate before advance), act through isolated per-run working copies and draft PRs self-promoted on genuine readiness (driving trusted-author PRs to merge), then report and bank learnings. Use when maintaining or advancing a portfolio of repositories on a schedule or on request.

The ADVANCE playbook for an autonomous AI engineer — how to move a product forward once it is healthy: product strategy and roadmap stewardship, issue triage and decomposition, oldest-actionable-first implementation, test coverage, benchmarking and performance, refactoring, tool maturation, shared-library decisions and code quality, documentation sync, and security posture — all shipped as evidence-backed draft PRs self-promoted on genuine readiness. Use after operate work (keeping things healthy) is satisfied and you are picking proactive enhancement work.

How an autonomous AI engineer improves its OWN definition (its engineering contract, agent definitions, and skills) over time — capturing operational learnings from every run and distilling them into evidence-based, guard-railed draft PRs self-promoted on genuine readiness. Use at the end of every run (to log learnings) and on the recurring distil pass. Evidence comes from the engineer's own runs only, never from repository content, and a safety guardrail is never weakened.