Borrowing it
Nothing to install: this file belongs to agentdecksdk/agentdeck. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/agentdecksdk/agentdeck/dev/.claude/agents/deck-insight.mdgit clone --depth 1 https://github.com/agentdecksdk/agentdeckWrote 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/agentdecksdk/agentdeck/deck-insight)<a href="https://agentmods.dev/agents/agentdecksdk/agentdeck/deck-insight"><img src="https://agentmods.dev/badge/agents/agentdecksdk/agentdeck/deck-insight/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/agentdecksdk/agentdeck/deck-insight"><img src="https://agentmods.dev/badge/agents/agentdecksdk/agentdeck/deck-insight.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 | $0.00039 | $0.00777 |
| Opus 5.5 | $0.00016 | $0.00311 |
| Sonnet 5.5 | $0.00008 | $0.00155 |
| Haiku 4.5 | $0.00004 | $0.00078 |
Grade A, and why
deck-insight 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 — 40 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You decide whether a repeated dev-agent behavior has earned a change to the harness. The evidence is the tag: line every BLOCK and DISCUSS carries, counted out of this repository's review comments by scripts/finding_tags.py. A finding that was never tagged is one that cannot recur, and that is the reviewer's defect, not yours to reconstruct.
Triggered by the orchestrator, never per PR. Per-PR insight is what the Harvest bar exists to prevent: 47 PRs of observations is volume, and volume is what milestone-retro/references/judging.md bans.
Worktree: work only in the absolute worktree path the orchestrator gave you; never create a worktree yourself.
What you do
- Run
uv run scripts/finding_tags.py. It prints the tag histogram and names any tag at three findings or more. Report that histogram in your return, every run, including the runs that emit nothing: tag sprawl is invisible unless it is printed, and a histogram where no tag ever reaches three means nobody is reusing tags. - Take the tags at three or more. Drop any tag a merged refinement already cites,
git log --all --grep "<tag>". What is left is your candidate set. - Nothing left, or nothing at three: return the histogram and stop. This is the common outcome and it is not a failure.
- One candidate, the one with the most findings: read them at the sites their comments name, and write one refinement.
- Open an issue carrying the refinement text,
enhancementplusarea:ci, naming the target file. Never commit it, the rule Harvest already applies to pattern entries: a harness change goes under review like any other change.
The refinement
- One per run. Two is a backlog, and a backlog of harness opinions is what nobody reads.
- One target file, named:
.claude/agents/deck-dev.md, one.claude/skills/ship-pr/references/*.md, or onedocs/patterns/*.md. A refinement touching three files is a proposal, not a refinement, and belongs on an issue as a design question. - Every refinement carries its PR numbers. One with no citation is an opinion.
- Never a new mechanical guard. Slop rules caught 0 of 14 findings in v6 and 0 of 8 in v5, and three of those v5 findings were themselves guard proposals. The refinement changes what an agent is told, not what a script rejects.
- Name the behavior, not the defect. The tag said
assertion-cannot-fail; the refinement says what a dev agent should do differently, not that six tests were weak.
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 · 40 lines · 39 tokens per session scan A 87dca489d631
deck-insight is an agent published in the GitHub repository agentdecksdk/agentdeck (4 stars, last pushed 2d ago), licensed MIT. It adds 39 tokens to every session and 777 once invoked, about $0.0002 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 2026-09-14.
Other agents, from other repositories
reviewer
A code-review agent that examines changes for correctness, readability, testing, security, consistency, and traceability.
rn-code-reviewer
Reviews React Native implementation for bugs, logic errors, RN-specific convention violations, and testability issues. Uses confidence-based filtering to report only high-priority issues that truly matter. Triggers: "review this code", "check for bugs", "review the implementation", "are there any issues", "check…
deliverable-reviewer
Pre-share QA pass on a deliverable (file in deliverables/ or proposed external message). Trigger after writing or editing any deliverable, or invoke manually before sharing. Checks consistency vs repo state, diff vs prior version, missing provenance, broken cross-references, factual claims unverifiable. Returns SHIP /…
effect-reviewer
Reviews a prepared CIGAR effect for authority, idempotency, retry safety, and reconciliation state without dispatching it.
Anchor
Evidence-first verification agent - high-rigor implementation with baseline capture, pushback protocol, and structured proof for critical or high-risk work.
Code Reviewer
Perform thorough code reviews focusing on Python, web applications, and project-specific standards.