tvproductions/gzkit / .agents/skills/gz-context-diet
gz-context-diet
VerifiedUse when the operator asks for a "diet" or "progressive disclosure" pass on AGENTS.md, CLAUDE.md or the rules.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 3.9 KB |
SKILL.md
View on GitHub---
name: gz-context-diet
persona: main-session
description: Use when the operator asks for a "diet" or "progressive disclosure" pass on AGENTS.md, CLAUDE.md or the rules.
category: code-quality
lifecycle_state: active
owner: gzkit-governance
last_reviewed: 2026-10-17
gz_command: chores show instructions-files-diet
metadata:
skill-version: "1.1.4"
model: sonnet
---
# gz-context-diet
## Purpose
Trigger-discovery wrapper for the `instructions-files-diet` chore. The
chore is the source of truth for the procedure, acceptance criteria, and
common rationalizations; this skill exists so the skill catalogue
pattern-matches on operator language ("diet," "progressive disclosure,"
"trim contract weight") that the generic `gz-chore-runner` description
does not surface.
This is **not** a substitute for reading the chore. The agent must read
`CHORE.md` before acting; the skill body below only routes there.
## Trigger
- **Binding (operator standing rule, 2026-05-29):** any budget-lift or
prose-trim edit to `AGENTS.md`, `CLAUDE.md`, or `.claude/rules/**` —
including lifting a bullet's rationale to `docs/governance/` to make
char-budget room for a new rule — routes through this skill, never a
direct `Edit`/`Write` to those files. Hand-editing instructions-file
prose to manage budget or shape is the bypass this skill exists to prevent.
- Operator asks for a "diet," "trim," or "progressive disclosure" pass on
the agent contract or memory files
- `uv run gz validate --instructions-files-budget` reports the
surface-delivery witness closing on a vendor cap (the advisory prints
remaining headroom in bytes); low headroom is the signal, not line count
- The advisory scorecard (`uv run gz validate --advisory-scorecard`)
surfaces Judgment-class bullets that duplicate Mechanical neighbors
## Procedure
Read the chore, then follow it:
```bash
uv run gz chores show instructions-files-diet
```
The Workflow, Acceptance Criteria, Anti-patterns, and Related sections
in that `CHORE.md` are the binding procedure. After reading, run the
chore through the standard chore lifecycle (the same five steps
`gz-chore-runner` enforces):
```bash
uv run gz chores plan instructions-files-diet
uv run gz chores advise instructions-files-diet
# ... apply edits per CHORE.md § Workflow ...
uv run gz chores run instructions-files-diet
uv run gz chores audit --slug instructions-files-diet
```
## Constraints (load-bearing reminders)
These three constraints justify a dedicated skill over generic
`gz-chore-runner` discovery — they are the points an agent must hold in
the per-turn surface even before reading `CHORE.md`:
1. **Narrative trim, never invariant relaxation.** *"Lighter ceremony is
not a tradeoff axis"* — `AGENTS.md` § MAKE LLM STOCHASTIC VIBES INERT
operative claim 1. Every Mechanical / Promotable bullet on the advisory
scorecard remains in the per-turn contract.
2. **Edit canonical, let sync propagate.** `.claude/rules/**` is
generated from `.gzkit/rules/**`. Direct mirror edits get overwritten
on the next `gz agent sync control-surfaces`.
3. **Foundation-kind content rigor regardless of lane label.** The chore
is lite-lane for execution overhead; the *content* (the agent
contract) is foundation-kind invariant surface. ARB receipts and
bullet-retention audit are required even though the lane is lite.
## Related
- Chore: `.gzkit/chores/instructions-files-diet/CHORE.md` (authoring source);
`src/gzkit/chores/instructions-files-diet/CHORE.md` is the generated package copy
- `gz-chore-runner` — generic chore-execution skill; this skill is the
trigger-discoverable specialization
- `gz-agent-sync`, `gz-arb`, `gz-validate` — wielded inside the chore
workflow
- `docs/governance/agent-contract-rationale.md` — the standing lift
target and precedent (six sections already lifted there)
- `docs/governance/advisory-rules-audit.md` — scorecard catalogue
- GHI #327 — origin
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |