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 babyworm/rtl-agent-team --skill p1-spec-research-policygit clone --depth 1 https://github.com/babyworm/rtl-agent-teamWrote 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/babyworm/rtl-agent-team/p1-spec-research-policy)<a href="https://agentmods.dev/skills/babyworm/rtl-agent-team/p1-spec-research-policy"><img src="https://agentmods.dev/badge/skills/babyworm/rtl-agent-team/p1-spec-research-policy/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/babyworm/rtl-agent-team/p1-spec-research-policy"><img src="https://agentmods.dev/badge/skills/babyworm/rtl-agent-team/p1-spec-research-policy.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, 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 152 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 | $0.00023 | $0.04095 |
| Opus 5.5 | $0.00009 | $0.01638 |
| Sonnet 5.5 | $0.00005 | $0.00819 |
| Haiku 4.5 | $0.00002 | $0.00409 |
Grade A, and why
p1-spec-research-policy 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 — 309 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Phase 1 Research Policy
Core Principles
- AskUserQuestion-First: Resolve every ambiguity, design choice, or scope decision via AskUserQuestion BEFORE proceeding. Never assume — a wrong assumption cascades to all later phases.
- Structured Interview (before spec parsing): Interview the user first — one question per
message covering goal, scope, constraints, priority trade-offs, verification strategy, and
dependencies (full interview protocol owned by the goal-clarifier agent). Record answers in
docs/phase-1-research/design-intent.md; resolve ambiguous spec language with the user's stated intent, never agent assumptions. - Approach Comparison for Open Items: When the spec allows multiple implementation paths,
present a per-OPEN comparison table (Approach | Pros | Cons | Area Est. | Latency Est. |
Recommendation) and let the user select via AskUserQuestion. Record choice + rationale in
open-requirements.json
resolution_rationale. - Incremental Requirement Approval: Seek approval in stages — interface/IO → functional (per block) → performance → open items — so users correct misinterpretations before they propagate. Finalize iron-requirements.json only after all stages are approved.
- Domain-Consult-First: Invoke domain-consult for algorithms, standards, coding tools, and HW trade-offs — never research in isolation. Experts provide knowledge; spec-analyst captures it as structured artifacts.
- Propose, Do Not Decide: Present algorithm/tool candidates with trade-offs; the user makes final selections. Architecture-level decisions (pipeline, partitioning, memory hierarchy) belong to Phase 2 — Phase 1 surveys and recommends.
- Exhaustive Tree Exploration: Spawn maximum parallel agents to explore every feasible solution path before committing. Skip ONLY if the user specifies exact algorithm + architecture (even then, explore at least 2 variants for validation).
Spec Refinement Criteria
AskUserQuestion MUST cover these areas (skip items already provided by user):
- Target codec, profile, level (e.g., H.264 High Profile Level 4.1)
- Target resolution and framerate (e.g., 1080p@60fps, 4K@30fps)
- Encoder, decoder, or both
- Interface protocol (AXI4, AXI4-Lite, APB, custom)
- Clock frequency target and process node (ASIC vs FPGA)
- Feature scope restrictions (e.g., "TQ only", "intra-only")
- Priority trade-off preference (throughput vs area vs power vs quality)
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 · 309 lines · 23 tokens per session scan A 19b6d14af177
p1-spec-research-policy is a skill published in the GitHub repository babyworm/rtl-agent-team (61 stars, last pushed 21d ago), licensed MIT. It adds 23 tokens to every session and 4,095 once invoked, about $0.0001 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
hardware-health-check
Out-of-band hardware health via Redfish BMC (read-only) — power state, component health, thermal and power readings, firmware inventory, SEL log triage. Use when determining whether a host is powered off versus unreachable, checking hardware faults, reviewing thermal or PSU state, or triaging BMC event logs.
hardware-saas-gtm
🇺🇸 Hardware + SaaS GTM Playbook — End-to-end go-to-market for consumer hardware products with software subscription models. Covers crowdfunding engineering (Kickstarter/Indiegogo/Makuake), KOL matrix for physical products, hardware PMF validation metrics, device-as-funnel subscription design, localization entry…
readable-verilog-generator
Use when creating, writing, reviewing, annotating, repairing, refactoring, or validating readable Verilog RTL, including synthesizable Verilog-2001 .v files, existing-RTL analysis, semantic comment annotation, testbench scaffold planning, ASIC-quality review, local or remote Vivado/xsim validation, evidence-backed…
electrum
Run the 5-phase product definition workflow for software-augmented hardware: Explore → High-Level Design → Component Arrangement → System Description → Gate Checklist, every document bundled into one HTML page. Use when the user has a hardware+software product idea to develop, or an existing spec, brief or PRD (txt…
Bring up an owned Yandex speaker for an agent
Assess and set up an owned Yandex speaker as an agent audio terminal: model/firmware compatibility, bounded root bootstrap, persistent maintenance, private backup and intentional isolation. Use for device bring-up or its recovery, not ordinary desktop commands.
step-parts
Find, evaluate, and download low-level common standard CAD parts from step.parts, such as screws, bolts, nuts, washers, bearings, standoffs, electronics parts, motors, connectors, and other off-the-shelf components. Use when Codex needs to search the hosted step.parts catalog, resolve fuzzy part names, standards…