Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/xg-gh-25/SwarmAInpx agentmods add skills/xg-gh-25/swarmai/s_persistWrote 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/xg-gh-25/swarmai/s_persist)<a href="https://agentmods.dev/skills/xg-gh-25/swarmai/s_persist"><img src="https://agentmods.dev/badge/skills/xg-gh-25/swarmai/s_persist/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/xg-gh-25/swarmai/s_persist"><img src="https://agentmods.dev/badge/skills/xg-gh-25/swarmai/s_persist.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 3 findings, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high YARA Match · line 3 YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).Fix: Remove offensive tool references and exploit code. Legitimate agent skills should not contain penetration testing tools, exploit frameworks, or reconnaissance utilities.
- high Prompt Injection · line 190 Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.Fix: Audit all comments and invisible characters. Remove any instructions that direct the agent to perform unauthorized actions. Use plain, reviewable content.
- medium Memory Poisoning · line 102 Skill injects content designed to persist in agent memory or context across interactions. Persistent injection can alter agent behavior long after the initial interaction.Fix: Do not allow untrusted input to persist in agent memory or context. Validate all content before storing and implement memory isolation between sessions.
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 | $0.00112 | $0.04236 |
| Opus 5.5 | $0.00045 | $0.01694 |
| Sonnet 5.5 | $0.00022 | $0.00847 |
| Haiku 4.5 | $0.00011 | $0.00424 |
Grade A, and why
persist 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.
How it starts
The opening of the file, as written. The whole thing — 235 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Persist — Unified Knowledge Routing
Persist knowledge to the correct destination. Routes content based on type and project context to the appropriate file and section. Replaces the old s_save-memory (which wrote everything to MEMORY.md).
Routing Decision Tree (3 steps)
Step 0: Should this be persisted AT ALL? (admission gate — default is SKIP)
Persisting is not free. A stored fact that drifts becomes a liability: stale → it misleads; fresh → it costs upkeep; and if it never drove a judgment, it was worthless the moment it was written. The best fix for a misleading stale entry is to never store it. Most passing facts should NOT be persisted. (AGENT R30#4 / P6.)
This is JUDGMENT, not a gate. No regex or hook can decide it — the same string (
91K) is a liability as a usage snapshot and fine as a budget cap. It cannot be mechanically enforced, so it lives here as a norm you must apply, not check off. Reading this as a denylist is the failure mode: don't pattern-match the examples below, ask the one question.
THE ONE QUESTION (apply to every number / status / fact before storing it):
Is this value dynamic — AND does it change no future judgment? If both → REJECT. A value earns a place ONLY if it's decision-relevant (it will change how the agent or user judges/acts later) AND stable (it won't be wrong next week without anyone touching it). Both must hold. The conjunction is the whole test:
- dynamic + drives-no-judgment → REJECT (a token count, a star snapshot).
- dynamic + decision-relevant → keep, but store the method not the frozen value (a version bump matters, so record "run
Xto get it", never the number itself).- stable + decision-relevant → keep (a convention, an architecture decision, a failure lesson, a principle).
- stable + drives-no-judgment → still SKIP (trivia is not knowledge).
Examples of REJECT (illustrations of the question, NOT a checklist to match):
- Continuously-drifting counts: LOC, file/test/skill counts, line numbers, star/fork snapshots, token sizes, "N sessions / N commits", percentages-of-the-moment.
- A raw number whose only use is "to know the number" — that's a measurement, taken live.
- Status true only right now: "daemon is up", "CI is green", "3 tabs open".
- One-off transient context with no cross-session reuse.
What ships with it
3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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 Changed · -1 lines 2bef00400c72
- 1mo ago First seen · 236 lines · 112 tokens per session scan A ffd5e0cece8f
persist is a skill published in the GitHub repository xg-gh-25/SwarmAI (54 stars, last pushed yesterday), licensed MIT. It adds 112 tokens to every session and 4,236 once invoked, about $0.0004 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 2026-08-30.
Other skills, from other repositories
why
Explain the provenance, authority, expiry, degradation state, and token accounting of CIGAR context already presented in this session.
checkpoint
Create an inspectable CIGAR checkpoint before compaction, interruption, or a meaningful task boundary.
remember
Routes user requests containing "remember", "recall", "checkpoint", "session", "todo", or "where were we" to the correct OpenEmpiric (OEM) MCP tool. Use when the user wants to persist, retrieve, or contextualize knowledge from project memory.
audit-knowledge
Scan Antigravity conversation transcript + artifact directory for extractable knowledge. Use when user asks for 'knowledge audit', 'audit knowledge', 'check for extractable knowledge', 'scan transcript', or at session start when audit cadence is exceeded.
handoff
Generate a passoff package so the next reader can pick up cleanly — future-you in a new session (typically when context is high) or a coworker. Default + auto modes emit a paste-ready next-session opener plus PROGRESS / CLAUDE / memory updates, commit, and /extract; brief emits an 80-150 word coworker brief (no file…
audit-share
Internal facet of the audit family — invoke via '/audit share'. Batch-reviews personal knowledge for promotion to the team-shared project-knowledge/ tier; per-item opt-in, requires the shared-knowledge config.