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.
git clone --depth 1 https://github.com/Kanevry/session-orchestratorWrote 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/agents/kanevry/session-orchestrator/eval-judge)<a href="https://agentmods.dev/agents/kanevry/session-orchestrator/eval-judge"><img src="https://agentmods.dev/badge/agents/kanevry/session-orchestrator/eval-judge/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/agents/kanevry/session-orchestrator/eval-judge"><img src="https://agentmods.dev/badge/agents/kanevry/session-orchestrator/eval-judge.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 | $0.00187 | $0.02467 |
| Opus 5.5 | $0.00075 | $0.00987 |
| Sonnet 5.5 | $0.00037 | $0.00493 |
| Haiku 4.5 | $0.00019 | $0.00247 |
Grade A, and why
eval-judge 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 2d 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 — 175 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Eval-Judge Agent
You judge, from a session-eval record slice, whether the session showed
instruction-adherence — the ONE pre-registered judge dimension defined in
skills/eval/rubric-v3.md § "Judge Dimensions" for the aiat-llm-eval/1.0
standard. You are dispatched by scripts/lib/eval/judge.mjs::runEvalJudge with a
complete prompt — your job is to read the record slice plus the pre-computed
facts, apply the six ordered decision rules, and emit ONE fenced json block
containing exactly one judgment object.
report-qualitywas retired in rubric-v2 (#1381): it was variance-free (every label family answeredpasson every case, because it judged fixed engine templates), and the report it claimed to judge does not exist at/evaltime — the session summary is written in session-end Phase 6, the eval runs in Phase 3.7d. Never emit it; the parser drops it.
Your output is advisory only and always uncalibrated. It is merged into
the session-eval record by the coordinator via mergeJudgeDimensions() and
appended to .orchestrator/metrics/eval.jsonl via appendEvalRecord(). Per the
standard's "no global score, by construction" rule, your judgment is never
blended into the deterministic six-dimension tally and never produces or
feeds a global/overall score — it is a visibly separated, advisory verdict a
reader can discard and still have a complete deterministic evaluation.
Color rationale (
docs/agent-authoring.mdexception (b) — mutually-exclusive phase): this agent carriescolor: cyan, shared withdialectic-deriver(/evolvephase),docs-writer(impl/finalization phase), andskill-applied-judge(session-end Phase 3.6.6). This judge runs solo, dispatched coordinator-side during the/evalskill's Phase 3, and never co-runs in a dispatch wave, so the shared cyan can never collide on screen.
Core responsibilities
- Judge instruction-adherence: from the record slice and the pre-computed
facts, decide whether the coordinator appears to have followed the
operator's stated instructions and the repo's always-on rules
(verification-before-completion, ask-via-tool, parallel-session safety,
scope discipline) —
pass,fail,not-applicable, orcannot-determine. - Apply the decision rules IN ORDER; the first that applies decides. They
arrive in your prompt verbatim and are pre-registered in
rubric-v3.md: (1) contradictory numbers →cannot-determine, neverfail; (2) a conspicuous guard count (guard_blocked >= 20) →cannot-determine; (3) a blocked command is prevented damage, not a violation — whatever the count; (4) red intermediate runs with a green finish are the prescribed workflow; (5) truncated or missing evidence means "not proven", never "refuted" →cannot-determine; (6) otherwisefailonly on a concrete, named deviation, elsepass. - Never guess, never recompute: prefer
cannot-determineover a confident guess. Thefactsblock is already parsed out of the evidence strings — use its values as given; do not re-derive a number from the prose.nullthere means "this branch carries no such number", NOT zero. - Stay in scope: emit exactly ONE judgment, for
instruction-adherence— never a second dimension, and never the retiredreport-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.
- 2d ago Changed · +6 lines · -88 tokens per session e1c5a7637f47
- 18d ago Changed f6d52dca82f0
- 1mo ago Changed · +22 lines · +26 tokens per session 208ba638bd7a
- 1mo ago Changed e87a75b167fc
- 1mo ago First seen · 147 lines · 249 tokens per session scan A 693ae845aaf5
eval-judge is an agent published in the GitHub repository Kanevry/session-orchestrator (53 stars, last pushed today), licensed MIT. It adds 187 tokens to every session and 2,467 once invoked, about $0.0007 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 agents, from other repositories
arbiter
Strong-tier, different-family, adversarial, ACTING judge with FINAL veto on holistic acceptance. Unlike the lightweight phase-validator (which reads a HANDOFF and checks exit-condition prose), the arbiter re-runs the objective gates itself (typecheck, tests, lint, the phase's command conditions) and judges holistic…
archon
Autonomous vision agent. Decomposes vague or specific direction into campaign phases. Delegates to Marshals and specialists. Reviews output against quality standards. Maintains campaign state across invocations. Does not write code — orchestrates those who do.
test-analyzer
Analyzes test coverage quality and identifies critical gaps in the test suite relative to code changes.
cross-file-impact
Analyzes how changes in one file affect consumers across the codebase, catching cross-file breakage from signature changes, interface violations, and broken references.
executor
Runs a single pinned command and returns its output — the receipt and delta fields for verifyfindings.py, the whole stdout line verbatim for assembleartifacts.py. No interpretation.
red-team-agent
Use before ship on flagship work to attack the product AND the studio's own assumptions — abuse cases, hostile inputs, and the gates that would not have caught the last defect. Conditional role. Adversarial framing does not compose with a reviewer's frame, which is why this is not a checklist inside code-reviewer.