Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add randomittin/heimdall --skill mac-deep-cleangit clone --depth 1 https://github.com/randomittin/heimdallWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/randomittin/heimdall/mac-deep-clean)<a href="https://agentmods.dev/skills/randomittin/heimdall/mac-deep-clean"><img src="https://agentmods.dev/badge/skills/randomittin/heimdall/mac-deep-clean/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/randomittin/heimdall/mac-deep-clean"><img src="https://agentmods.dev/badge/skills/randomittin/heimdall/mac-deep-clean.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00091 | $0.02945 |
| Opus 5.5 | $0.00036 | $0.01178 |
| Sonnet 5.5 | $0.00018 | $0.00589 |
| Haiku 4.5 | $0.00009 | $0.00295 |
Grade D, and why
mac-deep-clean scanned grade D with 3 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Asks for rootmediumPrivilege escalation
A mod that escalates privileges can change anything on the machine, not only the project.
- **`sudo -n true`** to test for cached sudo without prompting; if it fails, print the manual command instead of hanging. Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
find ~/.claude/projects -name '*.jsonl' -mtime +30 -print # list first; append -delete only after the user OKs the list Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- **Scope precisely.** `rm -rf` explicit paths only. Never `rm -rf` a parent when you mean a child. Verify exact version-dir names (Android SDK, NDK) before deleting "old" versions. How it starts
The opening of the file, as written. The whole thing — 192 lines — stays where its author put it; the contents beside it link to each section on GitHub.
mac-deep-clean
Autonomous macOS storage + memory cleanup. Investigate first (read-only), present a tiered reclaimable map, delete only what's regenerable or user-approved. Destructive ops are irreversible — measure, confirm, then act.
Ships with Heimdall: invoke it as hmd:mac-deep-clean (a copy you installed by hand under ~/.claude/skills/mac-deep-clean/ is plain mac-deep-clean). hmd:system-health and heimdall-cleanup --deep notice the need and hand off here — they never delete anything outside Heimdall's own garbage; this skill is the confirm-gated executor.
Golden rules
- Measure before deleting. Always
du -sh/df -hbefore and after. Report reclaimed bytes. - Irreversible = confirm. Deleting repos/files with uncommitted work needs explicit user naming. Never infer consent from background-task notifications.
- Scope precisely.
rm -rfexplicit paths only. Neverrm -rfa parent when you mean a child. Verify exact version-dir names (Android SDK, NDK) before deleting "old" versions. - Caches regenerate; user data does not. Keep: iMessage db, browser profiles, Notes, VSCode, source
.git, loose documents. Delete: caches, build output, node_modules, old SDK versions, dead worktrees. - Delegate investigation to parallel read-only agents. One agent per domain (repos / user Library / System Data). Spawn concurrently,
run_in_background: true.
macOS gotchas
timeoutdoes not exist on macOS. Userun_in_background: trueor a background&+ poll instead oftimeout N cmd.df -h /lies./is the sealed read-only system snapshot (a small, fixed-size volume). Real data lives on/System/Volumes/Data. Usediskutil info /→ "Container Free Space" for true free.- "System Data" bucket = swap + Preboot + caches + logs + diagnostics + purgeable. Not a folder — an aggregate.
ps auxRSS undercounts (shared/compressed pages). For real memory usetop -l 1 -o memandvm_stat. Numbers invm_statend in.— strip withsed 's/\.//'before math.sudo -n trueto test for cached sudo without prompting; if it fails, print the manual command instead of hanging.pnpm store prune/go clean -cachecan hang — preferrm -rf ~/Library/Caches/{pnpm,go-build}directly.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 7d ago First seen · 192 lines · 91 tokens per session scan D 6059421301c6
mac-deep-clean is a skill published in the GitHub repository randomittin/heimdall (5 stars, last pushed 3d ago), licensed MIT. It adds 91 tokens to every session and 2,945 once invoked, about $0.0004 per session on Opus 5.5. A static security scan graded it D with 3 findings (asks for root, reads agent configuration directories, recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-09.
Other skills, from other repositories
atomic-debug
Hypothesis-driven debugging skill. Use when a bug, test failure, crash, or unexpected behavior is reported. Auto-trigger on error pastes or "broken/doesn't work/failing" language. Explicit invocation: /atomic-debug. Output: symptom statement → hypothesis table → cheapest test first → root cause. No symptom-patching.…
sipcode-regress
Check the user's own local Claude Code and Codex logs for regressions after an update (prompt-cache misses, 5-minute cache TTL, uncached context, hourly cache rebuilds, Codex usage limit draining faster per token) and pin each one to the version that introduced it, using the Sipcode CLI. Use when the user asks whether…
vc:scout
Fast codebase scouting using shell search and optional parallel research agents. Use for file discovery, task context gathering, and quick scoped searches across directories.
vc:debug
Debug systematically with root-cause analysis before fixes. Use for bugs, test failures, unexpected behavior, performance issues, CI failures, or system investigation.
vc:chrome-devtools
Automate browsers with Puppeteer CLI scripts and persistent sessions. Use for screenshots, performance analysis, network monitoring, web scraping, form automation, JavaScript debugging.
vc:sequential-thinking
Apply step-by-step analysis for complex problems with revision capability. Use for multi-step reasoning, hypothesis verification, adaptive planning, problem decomposition, course correction.