stellar-raven: Skill for Claude Code

.agents/skills/golden-truth/SKILL.md

golden-truth is a skill for Claude Code, Codex from stellar-experimental/stellar-raven. It costs 71 tokens per session (4,370 once invoked), scanned A, original, Apache-2.0.

A verification workflow for changing the trusted question-and-answer set used to judge coding-agent evaluations. It checks important facts against multiple independent kinds of evidence before changing the record.

In plain words
What is it for?
Use it when editing evaluation questions, expected answers, key facts, exclusions, source references, freshness labels, or notes about disputed facts.
Why use it?
It prevents one mistaken source or unverified claim from becoming the standard used to judge every future evaluation.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one. Also seen: mentions subagents; mentions Claude Code; installed under .agents/ (shared by several agents).

This is stellar-experimental/stellar-raven's own configuration. It tells Claude Code and Codex how to work on stellar-raven itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything stellar-raven configures →

Reuse

Borrowing it

Nothing to install: this file belongs to stellar-experimental/stellar-raven. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/stellar-experimental/stellar-raven/main/.agents/skills/golden-truth/SKILL.md
Clone the repo
git clone --depth 1 https://github.com/stellar-experimental/stellar-raven

Made for: Claude Code, Codex.

Wrote 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.

agentmods badge for golden-truth

README.md
[![agentmods](https://agentmods.dev/badge/skills/stellar-experimental/stellar-raven/golden-truth/github.svg)](https://agentmods.dev/skills/stellar-experimental/stellar-raven/golden-truth)
Your own site
<a href="https://agentmods.dev/skills/stellar-experimental/stellar-raven/golden-truth"><img src="https://agentmods.dev/badge/skills/stellar-experimental/stellar-raven/golden-truth/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.

agentmods 80×15 button for golden-truth

Your own site · 80×15
<a href="https://agentmods.dev/skills/stellar-experimental/stellar-raven/golden-truth"><img src="https://agentmods.dev/badge/skills/stellar-experimental/stellar-raven/golden-truth.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 71 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 4,370 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector warn 7 Sept 2026
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 Rogue Agent · line 48
    Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.
    Fix: Remove any persistence mechanisms (cron jobs, startup scripts, state files). Skills should not maintain state across sessions without explicit user consent.
How audits are shown
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00071 $0.04370
Opus 5 $0.00036 $0.02185
Sonnet 5 $0.00014 $0.00874
Haiku 4.5 $0.00007 $0.00437

Measured 6d ago against content hash 200aa8759567, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-10, from the pricing page.

Security

Grade A, and why

golden-truth 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 6d 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.

.agents/skills/golden-truth/SKILL.md · 273 lines

How it starts

The opening of the file, as written. The whole thing — 273 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Golden gospel-truth verification — how to change the golden corpus without codifying lies

This skill is agent-agnostic: a plain-markdown runbook. Claude Code invokes it as a skill; Codex or any other CLI agent can be pointed at this file directly.

North star

The golden Q→A corpus is the gospel every eval round is judged against. A wrong agent answer costs one verdict; a wrong golden silently corrupts every future round, every A/B, every re-judge — and a golden "corrected" from thin evidence is worse than the error it replaced, because it now carries the authority of a review. Therefore:

Never change gospel from a single source class. Live-probing the corpus, or reading one docs page, is discovery — not verification. Gospel changes require multi-source triangulation across independent source classes, and disputed facts are encoded as disputes, never pinned.

This skill governs ANY change to judge-facing gospel — question, golden.answer, golden.keyFacts, golden.avoid, golden.notes, tags.freshness, tags.trap — and to the judge-blind truth block that carries its provenance. The corpus is owned: edits land directly in the per-case file eval/qa/corpus/battery/<category>/<id>.json, reviewed like code. The CI gospel-change lint (eval/qa/lint-corpus.mjs, diff-aware against the merge base) refuses any judge-facing change whose truth.verified did not change in the same diff with non-empty evidence + rootCause — that lint is the successor to the retired override-file enforcement, and it moves the check from compile-time bookkeeping to the moment of change.

The same evidence bar applies to a separately frozen provenance-bearing suite when its contract invokes this skill. Keep that suite in its contract-owned file. Do not compile it into the battery or existing routing lanes. Evaluation lanes never merge.

Step 1 — classify the truth domain (this picks the verification standard)

Domain What gospel means Examples
real-world / protocol Authoritative primary sources + source code. The corpus/aggregators may lag or be wrong — that's an improvements/ finding, never an excuse to weaken the golden. RPC limits, CLI commands, SEP semantics, build targets
corpus-grounded What the live community corpora (Scout, Lumenloop) support NOW, cross-checked against the real world. Distinguish real-world-confirmed from corpus-only (in the aggregator, no external footprint). Corpus-only facts may appear in goldens but must be labeled so graders treat them as source-relative. A corpus-vs-world contradiction → improvements/ finding + grade leniently on both sides. Regional community events, builder directories, project records
freshness-sensitive A behavior, never a pinned value. Point-in-time figures require an asOf date in the golden text itself, tags.freshness: "scheduled", and a truth.reverifyBy date. SCF amounts, country counts, versions, rosters

Read the full file on GitHub · 273 lines

Changes

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.

  1. 6d ago Changed · +2 lines 200aa8759567
  2. 9d ago First seen · 271 lines · 71 tokens per session scan A e8b1552226a1

Subscribe to this mod's changes

golden-truth is a skill published in the GitHub repository stellar-experimental/stellar-raven (6 stars, last pushed today), licensed Apache-2.0. It adds 71 tokens to every session and 4,370 once invoked, about $0.0004 per session on Opus 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 2026-08-31.

Related

Other skills, from other repositories

code-review

Code review assistance with linting, style checking, and best practices.

agno-agi/agno · 16 tokens

good-docs-audit

Audit a doc, guide, README, or block of prose against the good-docs-writing style guide and report violations. Use when the user asks to review, critique, lint, or check the voice and tone of documentation or text. Produces a structured findings report (file:line, rule violated, offending text, suggested rewrite) and…

ComposioHQ/composio · 82 tokens

cli-command

Design, implement, or review Composio CLI commands under ts/packages/cli using Effect, @effect/cli, services, output conventions, configuration and environment variables, and local vendor references. Use for CLI command UX, command wiring, service changes, or CLI source edits. Do not use for CLI E2E-only work; use…

ComposioHQ/composio · 77 tokens

docs-decisions

Work on Composio documentation content, Fumadocs configuration, changelogs, docs automation prompts, docs decisions, ADR-style records, and docs review guidance. Use for files under docs/, documentation workflows, or requests to record or update a docs decision. Do not use for SDK runtime changes unless docs are the…

ComposioHQ/composio · 72 tokens

cross-review

Verify an implementer's diff with an INDEPENDENT, different-vendor sub-agent (diff plus contract only); turn blocking issues into fix-tasks and loop until clean.

omnigent-ai/omnigent · 37 tokens

verify-implementation

A workflow that runs a project’s verification skills to produce a report on coding patterns, architecture rules, and project conventions. It is intended for work after implementation, before a pull request, or during code review.

sangrokjung/claude-forge · 37 tokens