feat: AI agent operations standard (K22-AGENT) and EX-0005 #6

2026-10-10T13:15:57Z

KofTwentyTwo commented

2026-10-10 13:15:57 +00:00

(Migrated from github.com)

What

standards/ai-agents.md — new K22-AGENT standard for how an agent session runs: instruction precedence, standards loaded fresh at every session start (including after context clear/compaction), autonomy boundaries, minimal-change and verification-honesty discipline, secrets/untrusted-content rules, provenance trailers per K22-AI-02, context economy. Complements policies/ai-assisted-development.md (governance) with session operations.
exceptions/register.md — EX-0005: agents push directly to the four Internal-tier personal-automation repos (second-brain, nix, Praetor develop, Jellyfin) where that is the designed workflow; reasons, risks, compensating controls, yearly review. K22-AGENT-04 requires agents to resolve such conflicts through the register instead of silently violating or refusing.
README.md — one row in the "If you want to…" table.

Notes for review

The K22-AGENT MUSTs are verified by review/configuration review (like the K22-AI policy controls), not by Test-RepoConformance.ps1 — agent session behavior isn't repo-checkable; no checker change included.
New MUSTs normally mean a MAJOR release; the repo is pre-1.0 draft, so this lands before the v1.0.0 baseline.
K22-AI-11 assessment behind EX-0005's "in part": the agent permission classifier does confirm destructive/protected operations (it blocks blind IaC applies); the gap is unconfirmed git push in the four listed repos only.
Downstream (nix repo, separate change): ~/.ai becomes a thin private layer (profile/preferences) over a standards clone that session-start hooks keep fresh — that work lands after this merges.

Generated with Claude Code

## What - **`standards/ai-agents.md`** — new K22-AGENT standard for how an agent session runs: instruction precedence, standards loaded fresh at every session start (including after context clear/compaction), autonomy boundaries, minimal-change and verification-honesty discipline, secrets/untrusted-content rules, provenance trailers per K22-AI-02, context economy. Complements `policies/ai-assisted-development.md` (governance) with session operations. - **`exceptions/register.md`** — EX-0005: agents push directly to the four Internal-tier personal-automation repos (second-brain, nix, Praetor develop, Jellyfin) where that is the designed workflow; reasons, risks, compensating controls, yearly review. K22-AGENT-04 requires agents to resolve such conflicts through the register instead of silently violating or refusing. - **`README.md`** — one row in the "If you want to…" table. ## Notes for review - The K22-AGENT MUSTs are verified by review/configuration review (like the K22-AI policy controls), not by `Test-RepoConformance.ps1` — agent session behavior isn't repo-checkable; no checker change included. - New MUSTs normally mean a MAJOR release; the repo is pre-1.0 draft, so this lands before the v1.0.0 baseline. - K22-AI-11 assessment behind EX-0005's "in part": the agent permission classifier does confirm destructive/protected operations (it blocks blind IaC applies); the gap is unconfirmed `git push` in the four listed repos only. - Downstream (nix repo, separate change): `~/.ai` becomes a thin private layer (profile/preferences) over a standards clone that session-start hooks keep fresh — that work lands after this merges. Generated with [Claude Code](https://claude.com/claude-code)

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: KofTwentyTwo/standards#6