AI Signal 562
Agent skills that bring team coding standards to Claude Code and Codex
ADLC Team Skills equips AI coding agents with a shared team constitution, architectural standards, and evaluation benchmarks so they operate as compliant, accountable members of an engineering group.
For engineers who build or operate software, this shifts AI agents from isolated guessers to tools that automatically load team-specific context at the start of each session, reducing guesswork and unreviewable pull requests. It also creates a closed feedback loop where agent-generated learnings are committed back to the repository, improving long-term code quality. Adoption requires installing the skill repo and maintaining a team-ai-directives folder, which adds setup effort but aims to cut technical debt and context rot.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
The skill injects a compact index of the team’s constitution and active decisions at session start, enabling progressive disclosure of relevant rules.
It enforces a contract-first workflow where mission-brief defines goals, constraints, non-goals and success criteria before any code is written.
It provides verification-first evaluation through automated LLM judges and binary graders that test code against business risks before human review.
THE READ
What elseif makes of it.
The material describes a common problem: individual prompt hacks give quick wins for solo developers but, when scaled, produce chaotic technical debt, context rot, unreviewable PRs and lost code ownership. The proposed solution is a shared cognitive layer that turns AI agents into compliant team members by loading the team’s constitution, product strategy, architectural standards and evaluation benchmarks at the start of each session.
Installation is done via the npx adlc-skills-cli tool, which adds the skill repo, generates slash commands and wires session_start event hooks for supported agents. Once installed, the team-boot hook auto-loads a ~100-token index of the team constitution and active decisions, and the mission-brief step defines a contract-first specification before code generation begins.
A closed feedback loop is created by the levelup-specify process, which extracts execution traces from agent sessions and commits new rules directly to the git-tracked directives repository. Verification is performed first by automated LLM judges and binary graders that assess code against business risks, only then passing it to human review.
Adoption requires setting up a team-ai-directives repository, running the team-setup command to scaffold or clone the directives, and maintaining the constitution, CDR, PDR/ADR files and skills manifest. The approach works only with agents that support the Agent Skills standard; if an agent lacks that support or the team fails to keep its directives current, the benefit diminishes and the agent may revert to guessing behavior.
Written by elseif from the cluster below · checked for specifics the sources never containedTHE CLUSTER