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 wrg32786/aigent-os --skill capsule-compactgit clone --depth 1 https://github.com/wrg32786/aigent-osWrote 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/wrg32786/aigent-os/capsule-compact)<a href="https://agentmods.dev/skills/wrg32786/aigent-os/capsule-compact"><img src="https://agentmods.dev/badge/skills/wrg32786/aigent-os/capsule-compact/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/wrg32786/aigent-os/capsule-compact"><img src="https://agentmods.dev/badge/skills/wrg32786/aigent-os/capsule-compact.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 | $0.00042 | $0.00464 |
| Opus 5.5 | $0.00017 | $0.00186 |
| Sonnet 5.5 | $0.00008 | $0.00093 |
| Haiku 4.5 | $0.00004 | $0.00046 |
Grade A, and why
capsule-compact scanned grade A with 0 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 1mo 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
/capsule-compact
Run python3 daemons/capsule-compact.py <head_capsule_id> from the vault root.
The script:
- Walks
parent_capsule_idchain backward from the head - If chain length >= 5, summarizes the oldest 3 capsules into a single
chain-summary-<head>-<date>.mdcapsule - Updates pointers: boundary capsule's parent → summary; summary's parent → original chain root's parent
- Marks each compacted capsule with
compacted_into: <summary_id>(they stay on disk for archeology)
When to run
- /close Step 0.5 PART E (v0.4.5) emits
[CADDY:context] CHAIN — capsule chain length Nwhen threshold hit - Manually before a session-heavy week to keep /open Step 2.5 fast
- After narrative-reconstruction archeology when the chain has grown unwieldy
Defaults
- Threshold: 5 capsules
- Summarize-count: 3 oldest
Override via --threshold N and --summarize-count N flags.
What it does NOT do
- Does not auto-compact at /close. Hints, you decide.
- Does not delete compacted capsules. Each gets
compacted_into:marker. - Does not LLM-summarize. Concatenates structured fields (objective, open_threads still live, held decisions).
- Does not walk forks. parent_capsule_id is strictly linear.
Cross-links
- [[concepts/Somatic v0.4.5 Chain Compaction]] — spec
- [[concepts/context-capsule]] — base capsule schema
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.
- 1mo ago First seen · 50 lines · 42 tokens per session scan A c92730d0c625
capsule-compact is a skill published in the GitHub repository wrg32786/aigent-os (23 stars, last pushed yesterday), licensed MIT. It adds 42 tokens to every session and 464 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-01.
Other skills, from other repositories
setup-brain
Set up or upgrade an AI-powered Obsidian vault. Interviews you, builds your vault structure (or works with what you already have), creates your CLAUDE.md memory file, installs tools, and gets you journaling — all in one conversation. Also has a repair/upgrade path for existing users.
second-brain-mapping
Use when the user wants to map or remap their second brain, refresh the vault's queryable metadata index, extract structured frontmatter from typed notes (books, meetings, people, journals, goals), fill wikilink gaps, or surface cross-type insights and patterns no single file shows. Triggers: /second-brain-mapping…
evolve
Use when many related hardened instincts (high-confidence feedback/discovery memories) have piled up in one domain, when the user runs /evolve or asks to promote, cluster, graduate, or consolidate accumulated instincts into a Command, Skill, or Agent, or when files in Instinct Proposals need review. Part of the…
vault-system
Use when maintaining the vault ITSELF as a system, not its content: knowledge-graph build/rebuild/query (graphify), second-brain mapping, vault diagnose/audit/hygiene, drift detection, extract rules from vault, backfill journals, setup vault types, consolidate memory (index over cap), Obsidian tooling…
diagnose
Use when an AI Brain vault needs a health check or something feels off: Claude ignores CLAUDE.md, journal entries missing from /weekly, hooks not firing, something broke after a git pull, a friend's install misbehaves, Obsidian crashes or pegs the CPU, a .claude/worktrees folder appeared, Granola/WhatsApp/Slack/Gmail…
nav-compact
Clear conversation context while preserving knowledge via context marker. Use when user says "clear context", "start fresh", "done with this task", or when approaching token limits.