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 EmpiricaAI/empirica --skill pre-action-groundinggit clone --depth 1 https://github.com/EmpiricaAI/empiricaWrote 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/empiricaai/empirica/pre-action-grounding)<a href="https://agentmods.dev/skills/empiricaai/empirica/pre-action-grounding"><img src="https://agentmods.dev/badge/skills/empiricaai/empirica/pre-action-grounding/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/empiricaai/empirica/pre-action-grounding"><img src="https://agentmods.dev/badge/skills/empiricaai/empirica/pre-action-grounding.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 2 findings, up to medium
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 →
- medium Excessive Agency · line 47 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
- medium Excessive Agency · line 109 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
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 | $0.00132 | $0.01365 |
| Opus 5.5 | $0.00053 | $0.00546 |
| Sonnet 5.5 | $0.00026 | $0.00273 |
| Haiku 4.5 | $0.00013 | $0.00136 |
Grade A, and why
pre-action-grounding 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 — 119 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Pre-Action Grounding
Increase prediction confidence by investigating first and asking last — then fill the knowledge graph and go. The questioning part is minimal: enough to close what investigation cannot reach, never a substitute for it.
Why this exists, measured: across 59 practice databases, 3,011 of 5,484
success criteria were the auto-filled Goal completion achieved and 5,462
(99.6%) carried the same validation_method. Four goals in five said "done when
done." The asking was missing, and a default filled the slot so the absence read
as an answer.
The protocol
decompose → investigate → grade → ask (residue only) → bank → emit → go
1. Decompose
Break the ask into what must be true for "done" to be checkable. Not task steps — knowledge preconditions. For each: could I check this today, and with what?
2. Investigate (before any question)
One noetic-batch bundling the lot: project-search --task (prior goals,
findings, dead-ends on this topic), investigate, targeted reads, and — mesh
present — cortex search_knowledge. Most done-conditions are knowable; usually
only the requester's intent is not.
3. Grade
Reuse the CHECK claims vocabulary — do not invent a parallel one:
| Grade | Meaning |
|---|---|
ran |
executed and observed this session |
read |
opened the source this session |
retrieved |
from a prior artifact — testimony, not observation |
assumed |
acting without checking |
4. Ask — only the residue
A question is earned only by an item that is still assumed and
load-bearing and knowable only from the requester's intent. Typically 0–2.
The discriminator is the Sentinel's own: is this knowable from data I can pull,
or only from the requester's head?
5. Bank the rest
Ungrounded residue you did NOT ask about → assumption-log --confidence <0-1>,
edged to the goal. Never a silent guess. External material cited → source-add,
linked via sourced_from.
6. Emit
What ships with it
1 file 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 First seen · 119 lines · 132 tokens per session scan A f9e4c61cb1df
pre-action-grounding is a skill published in the GitHub repository EmpiricaAI/empirica (251 stars, last pushed yesterday), licensed MIT. It adds 132 tokens to every session and 1,365 once invoked, about $0.0005 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 2026-09-03.
Other skills, from other repositories
workflow-builder
Design, build, and maintain autonomous OpenClaw workflows (stewards). Use when creating new workflow agents, improving existing ones, evaluating automation opportunities, or debugging workflow reliability. Triggers on "build a workflow", "create a steward", "automate this process", "workflow audit", "what should I…
smart-delegation
Intelligent task delegation — route to think with deep reasoning for hard problems, or Grok for unfiltered takes. Teaches when to escalate, how to pack context into sub-agent spawns, and how to communicate delays transparently. Default: handle directly on chat (thinking off). Escalate only when the quality gain…
claude-code
Route real repo work to Claude Code instead of editing by hand. Triggers on "claude code" or "cc", and on any request to edit, fix, refactor, or open a PR in a repo outside /.openclaw/workspace. Claude Code picks up the repo's CLAUDE.md / AGENTS.md, applies its standards, and knows the /ai-coding-config:multi-review…
review
A pause before an artifact goes into the world. Reviews external comms, money, calendar, public posts, or send-as-operator actions through a small panel of independent lenses (empathy first) and returns a verdict of pass / edit / hold / block.
cortex
Personal knowledge compiler and memory system. Ingests raw sources (documents, notes, transcripts, captures) into a structured, interlinked knowledge base. Maintains living entity pages, learning analysis, and the MEMORY.md routing table. Replaces the librarian skill.
drive-to-done
Drive a task all the way to a verified done state — write DoD first, verify each item with evidence, stop only at named stop conditions.