skill-loadout-manager
Manages named skill profiles (loadouts) so you can switch between focused skill sets and prevent system prompt bloat from too many active skills.
git clone --depth 1 https://github.com/ArchieIndian/openclaw-superpowers /tmp/skill-loadout-manager && cp -r /tmp/skill-loadout-manager/skills/openclaw-native/skill-loadout-manager ~/.claude/skills/skill-loadout-managerSKILL.md
# Skill Loadout Manager ## What it does Installing more skills increases OpenClaw's system prompt size. Every installed skill contributes its description to the context window on every session start — even skills you haven't used in months. Skill Loadout Manager lets you define named loadouts: curated subsets of skills for specific contexts. You switch to a loadout and only those skills are active. Everything else is installed but dormant. Examples: - `coding` — tools for writing, testing, reviewing code - `research` — browsing, fact-checking, note synthesis - `ops` — monitoring, cron hygiene, spend tracking - `minimal` — just the essentials: memory, handoff, recovery ## When to invoke - When you notice system prompt bloat slowing context initialisation - When switching between focused work modes (deep coding vs. research) - When you want to test a single skill in isolation - After adding many new skills that aren't always relevant ## Loadout structure A loadout is a named list of skill names stored in state. Activating a loadout signals to OpenClaw's skill loader which skills to surface in the system prompt. Skills not in the active loadout remain installed but excluded from description injection. ```yaml # Example loadout definition name: coding skills: - systematic-debugging - test-driven-development - verification-before-completion - skill-doctor - dangerous-action-guard ``` ## How to use ```bash python3 loadout.py --list # Show all loadouts and active one python3 loadout.py --create coding # Create new loadout (interactive) python3 loadout.py --add coding skill-doctor # Add skill to loadout python3 loadout.py --remove coding skill-doctor # Remove skill python3 loadout.py --activate coding # Switch to loadout python3 loadout.py --activate --all # Activate all skills python3 loadout.py --show coding # List skills in a loadout python3 loadout.py --status # Current active loadout python3 loadout.py --estimate coding # Estimate token savings ``` ## Procedure **Step 1 — Assess current footprint** ```bash python3 loadout.py --estimate --all ``` This shows the estimated description token count for all installed skills and highlights candidates for loadout pruning. **Step 2 — Define your loadouts** Think in contexts: What skills do you actually need when writing code? When doing research? During maintenance windows? Create one loadout per context, aiming for 5–10 skills each. ```bash python3 loadout.py --create coding python3 loadout.py --add coding systematic-debugging test-driven-development python3 loadout.py --add coding verification-before-completion dangerous-action-guard ``` **Step 3 — Activate a loadout** ```bash python3 loadout.py --activate coding ``` OpenClaw reads the active loadout from state on next session start and only injects those skill descriptions. **Step 4 — Switch as needed** Switching is instant and takes effect on the next session. No restart required. **Step 5 — Return to full mode** ```bash python3 loadout.py --activate --all ``` ## State Active loadout name and all loadout definitions stored in `~/.openclaw/skill-state/skill-loadout-manager/state.yaml`. Fields: `active_loadout`, `loadouts` map, `switch_history`. ## Notes - Always-on skills (e.g., `dangerous-action-guard`, `prompt-injection-guard`) can be marked `pinned: true` so they're included in every loadout automatically. - The `minimal` loadout is pre-seeded at install time with only safety and recovery skills.
Syncs agent daily memory and MEMORY.md to an Obsidian vault so notes are human-browsable. Use nightly or on demand.
Structured ideation before any implementation. Use when starting any non-trivial task.
Scaffolds and validates new superpowers skills. Use when creating a new skill for this repository.
Executes plans task-by-task with verification. Use when implementing a plan.
Triggers a secondary verification pass for any agent output containing factual claims, numbers, dates, or named entities before the output is acted on
Crawls a new codebase to infer stack, conventions, and key invariants, then generates a PROJECT.md context file for the agent
Handles PR review feedback by fetching comments, grouping issues, fixing one group at a time, and verifying before replies.
Detects skill name shadowing and description-overlap conflicts that cause OpenClaw to trigger the wrong skill or silently ignore one when two skills compete for the same intent.