How do I install the Harden skill?

Run npx skillmds@latest add horizonbrute/harden in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Harden safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Harden?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Harden free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Harden?

HorizonBrute (@horizonbrute) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Harden by horizonbrute — Agent Skill

Name: Harden
Author: HorizonBrute

Harden

Apply the authoritative brains-group ACL model to the AIOS layer — grants brains Read+Execute on bin/skills_bin, a no-write Deny across $HORIZON_SYSTEM, and an explicit full Deny on sbin/skills_sbin/logs. Use when the user types /harden, asks to "harden the AIOS", "apply the ACL model", "lock down sbin", or after horizon_aios_doctor.py reports a missing Deny ACE.

HorizonBrute Updated Sep 17, 2026

File contents

Skill: /harden

Model preference: #highcap (per horizon_aios_model_prefs.md; overridable by a prompt directive).

Apply the authoritative brains-group ACL model to the AIOS layer (horizon_aios_harden.py), enforcing security_invariants.md §2/§3/§5. Run at bootstrap, after adding/removing directories under $HORIZON_SYSTEM, or when horizon_aios_doctor.py reports a missing brains Deny ACE.

When to invoke

The user types /harden, or asks to "harden the AIOS", "apply the ACL model", "lock down sbin/skills_sbin", or /doctor flagged a missing Deny ACE on a privileged dir.

Arguments

/harden [--strict] [--dry-run] [--horizon-root PATH]

--strict — additionally strip inherited ACEs (Windows /inheritance:r; Unix chown -R + mode bits) for a locked-down standalone install. Default (additive) preserves existing/infra ACLs and enforces the model by adding ACEs only.
--dry-run — print every action without executing anything (needs no elevation).
--horizon-root PATH — explicit root; otherwise derived from the script location.

Step-by-step execution

Step 1 — Check elevation

ACL changes require the administrative context (Administrator on Windows, root on Unix). If the session is not elevated and the user did NOT pass --dry-run, warn that ACL changes will fail and instruct them to re-run from an elevated terminal (or with sudo). Do not pretend it succeeded. --dry-run needs no elevation — run it directly to preview.

Step 2 — Run

python "$HORIZON_SYSTEM/sbin/horizon_aios_harden.py"

Append --strict / --dry-run / --horizon-root as the user requested. The script warns and continues (rather than crashing) when not elevated, so failures are visible.

Step 3 — Report

Relay the banner output and the final Summary block (the brains grant/deny model applied). If the script reports the brains group was unavailable, surface that only owner-side hardening ran and it must be re-run after the group exists.

Notes for the executing agent

Owner + SYSTEM + Administrators always retain Full control (Windows), regardless of mode — the script never strips them.
All Deny ACEs are applied AFTER all brains grants, so an inherited permission can never reach a privileged dir. Do not reorder anything — just run the script.
The model: bin/skills_bin → brains Read+Execute; sbin/skills_sbin/logs → brains DENY (full); rest of $HORIZON_SYSTEM → brains no-write.
Never set AIOS_SKIP_HARDEN=1 in native deployments — that bypasses the hardening (it exists only for the Docker root-context build step).

---name: hardendescription: Apply the authoritative brains-group ACL model to the AIOS layer — grants brains Read+Execute on bin/skills_bin, a no-write Deny across $HORIZON_SYSTEM, and an explicit full Deny on sbin/skills_sbin/logs. Use when the user types /harden, asks to "harden the AIOS", "apply the ACL model", "lock down sbin", or after horizon_aios_doctor.py reports a missing Deny ACE.---# Skill: /harden**Model preference:** `#highcap` (per `horizon_aios_model_prefs.md`; overridable by a prompt directive).Apply the authoritative brains-group ACL model to the AIOS layer (`horizon_aios_harden.py`), enforcing security_invariants.md §2/§3/§5. Run at bootstrap, after adding/removing directories under `$HORIZON_SYSTEM`, or when `horizon_aios_doctor.py` reports a missing brains Deny ACE.---## When to invokeThe user types `/harden`, or asks to "harden the AIOS", "apply the ACL model", "lock down sbin/skills_sbin", or `/doctor` flagged a missing Deny ACE on a privileged dir.---## Arguments`/harden [--strict] [--dry-run] [--horizon-root PATH]`- `--strict` — additionally strip inherited ACEs (Windows `/inheritance:r`; Unix `chown -R` + mode bits) for a locked-down standalone install. Default (additive) preserves existing/infra ACLs and enforces the model by adding ACEs only.- `--dry-run` — print every action without executing anything (needs no elevation).- `--horizon-root PATH` — explicit root; otherwise derived from the script location.---## Step-by-step execution### Step 1 — Check elevationACL changes require the **administrative context** (Administrator on Windows, root on Unix). If the session is not elevated and the user did NOT pass `--dry-run`, warn that ACL changes will fail and instruct them to re-run from an elevated terminal (or with `sudo`). Do not pretend it succeeded. `--dry-run` needs no elevation — run it directly to preview.### Step 2 — Run```python "$HORIZON_SYSTEM/sbin/horizon_aios_harden.py"```Append `--strict` / `--dry-run` / `--horizon-root` as the user requested. The script warns and continues (rather than crashing) when not elevated, so failures are visible.### Step 3 — ReportRelay the banner output and the final Summary block (the brains grant/deny model applied). If the script reports the `brains` group was unavailable, surface that only owner-side hardening ran and it must be re-run after the group exists.---## Notes for the executing agent- Owner + SYSTEM + Administrators always retain Full control (Windows), regardless of mode — the script never strips them.- All Deny ACEs are applied AFTER all brains grants, so an inherited permission can never reach a privileged dir. Do not reorder anything — just run the script.- The model: `bin`/`skills_bin` → brains Read+Execute; `sbin`/`skills_sbin`/`logs` → brains DENY (full); rest of `$HORIZON_SYSTEM` → brains no-write.- Never set `AIOS_SKIP_HARDEN=1` in native deployments — that bypasses the hardening (it exists only for the Docker root-context build step).

Frequently asked questions