feat: clean up agent config

modules/capabilities/ai/instructions/codex.md

@@ -1,23 +1,24 @@
 # Environment
 
-- This is Nix; use `nix run` for missing commands. Default Nix commands to `--quiet`.
-- Keep output bounded. Capture large Nix listings in `/tmp`; inspect targeted excerpts and failing logs only. Locate store files before reading them.
+- Use `nix run` for missing commands and `--quiet` for Nix commands.
+- Capture large Nix listings in `/tmp`; read targeted excerpts and failure logs. Locate store files before reading them.
 - Manage persistent binaries, skills, plugins and MCP definitions in `/home/kiri/.config/nixos`. Keep interactive settings, credentials and session state outside Nix.
-- `HERDR_ENV=1` means you are inside Herdr. Read the Herdr skill or run `herdr --skill` when Herdr interaction is needed.
 
 # Work
 
-- Inspect relevant code and callers; follow project conventions and fix root causes.
-- Prefer existing code, standard libraries and native capabilities. Avoid speculative abstractions; optimize conceptual simplicity, not line count.
-- Keep changes focused. Prefer clean breaking changes over compatibility layers or migration scripts; explain required manual steps.
-- Use targeted reads and tool calls. Verify unfamiliar or consequential APIs against version-appropriate primary documentation; prefer Context7 when needed.
-- In reviews, distinguish confirmed problems from optional improvements and explain meaningful tradeoffs.
+- Complete authorized work without repeated confirmation. Make reasonable assumptions for routine choices; ask about consequential unknowns while continuing independent work.
+- Prefer existing capabilities and simple solutions. Add abstractions or compatibility code only for a concrete need; explain required manual steps.
+- Verify unfamiliar or consequential APIs against version-appropriate primary documentation; prefer Context7.
 - Never push without my explicit approval. Other remote actions must remain within the authorized task.
 - User instructions take precedence over skill guidelines, including testing defaults.
 
 # Validation
 
-Do not add or run tests unless explicitly requested or necessary to resolve a
-concrete correctness concern. Use the smallest relevant validation and stop once
-it provides sufficient confidence. Avoid broad suites and repeated builds without
-a specific reason. Report only checks actually performed.
+- Write or run automated tests only when explicitly requested or covered by the repository's established testing policy. Do not invent a testing policy or introduce tests into a repository without tests unless explicitly requested.
+- Validate changes with the smallest relevant build, manual behaviour check, or review of code and output.
+- Broaden or repeat validation only for changes, failures or unresolved concerns. Report only checks performed.
+
+# Writing
+
+- Lead with the result. Write concisely in plain language, preserving important reasoning, evidence and limitations.
+- Omit repetition and stock phrases. Use lists when they improve readability.