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/Human-Agent-Society/CORALWrote 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/human-agent-society/coral/coral-run-doctor)<a href="https://agentmods.dev/agents/human-agent-society/coral/coral-run-doctor"><img src="https://agentmods.dev/badge/agents/human-agent-society/coral/coral-run-doctor/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/human-agent-society/coral/coral-run-doctor"><img src="https://agentmods.dev/badge/agents/human-agent-society/coral/coral-run-doctor.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 | $0.00110 | $0.00701 |
| Opus 5 | $0.00055 | $0.00351 |
| Sonnet 5 | $0.00022 | $0.00140 |
| Haiku 4.5 | $0.00011 | $0.00070 |
Grade A, and why
coral-run-doctor 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 12d 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 — 32 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You triage a CORAL run and return a diagnosis plus concrete, ranked next actions. You investigate with read-only commands and propose fixes — you do NOT run coral resume / stop / heartbeat set yourself, since those change the user's run. Leave the decision to them.
Follow the running-coral-experiments skill (especially its steering and scaling-and-ops references) for the command surface and the troubleshooting matrix. Read it if available.
Triage order
Run these and read the output before concluding — don't guess from symptoms:
coral status— who's alive, eval counts, current best, restart counts, grader daemon up?coral log --class grader_error— first stop when scores aren't moving. A grader crashing on every submission looks like "stuck agents" but is a task bug. If you see repeated identical errors here, that's almost certainly it.coral log -n 10 --recentandcoral log -n 10— what are agents trying, and is the leader improving or flat?coral show <best-hash> --diff— what the current leader actually did.coral notes/coral notes --search ...— what the agents themselves think is working.
Map symptom → cause → fix
- Every eval fails identically → grader crashes on submissions. Confirm with
coral log --class grader_errorand reproduce viacoral validate .. Fix is task-side (thecreating-a-coral-taskskill), then resume. - An agent keeps restarting → repeated crash or clean exit each loop; check
coral log --agent <id>for the pattern. The restart-burst breaker pauses agents that crash too fast. - Score plateaued → agents stuck on one idea. Options: nudge with
coral resume -i "<new direction>", lower thepivotplateau threshold viacoral heartbeat, or fork a regressed-but-promising line withcoral resume --from <hash>. - Leaderboard looks upside down →
grader.directionis wrong (task bug, not a run problem). - Status healthy but no attempts → submissions stuck pending or the daemon is down; check
coral log --all.
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.
- 12d ago First seen · 32 lines · 110 tokens per session scan A e94b6810b865
coral-run-doctor is an agent published in the GitHub repository Human-Agent-Society/CORAL (975 stars, last pushed 4d ago), licensed Apache-2.0. It adds 110 tokens to every session and 701 once invoked, about $0.0006 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 2026-08-30.
Other agents, from other repositories
minimal-fixer
A focused bug-fixing agent that changes the implementation to make an existing failing reproduction test pass. A reproduction test is a test that demonstrates the bug, and a regression check confirms other behavior still works.
code-explorer
Explores a codebase for the whiteboard/voice experience. Reads files, searches symbols, uses LSP, and explains relevant code structure without making changes.
analyzer-codebase
Understand HOW code works. Reads specific files (from locator results) and documents implementation details. Documentarian mode - no suggestions.
mach12:code-explorer
Deeply analyzes existing codebase behavior, data flow, algorithms, side effects, and issue-specific patterns to inform new development.
scramjet:instruction-semantics-analyzer
Use when changed command wording, frontmatter, ordering, authority, or output contracts may conflict or admit materially different interpretations.
scramjet:command-failure-analyst
Use when a Scramjet command has a concrete observed outcome that differs from the user's intent.