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 EmpiricaAI/empirica --skill agent-pipelinegit clone --depth 1 https://github.com/EmpiricaAI/empiricaWrote 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/empiricaai/empirica/agent-pipeline)<a href="https://agentmods.dev/skills/empiricaai/empirica/agent-pipeline"><img src="https://agentmods.dev/badge/skills/empiricaai/empirica/agent-pipeline/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/empiricaai/empirica/agent-pipeline"><img src="https://agentmods.dev/badge/skills/empiricaai/empirica/agent-pipeline.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 | $0.00083 | $0.01810 |
| Opus 5.5 | $0.00033 | $0.00724 |
| Sonnet 5.5 | $0.00017 | $0.00362 |
| Haiku 4.5 | $0.00008 | $0.00181 |
Grade A, and why
agent-pipeline 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 8d 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 — 100 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agent pipeline: map, tag, act, review
Reviewing code and fixing it are different jobs with different costs. This pipeline puts each stage on the cheapest tier that its output can be mechanically checked at, and refuses to let an unchecked output become work for the next stage.
map (sonnet) -> tag (opus) -> verify (skeptic + the tagger's own test) -> act (haiku, sandboxed) -> review (agent-review)
pointers findings, default-refute, planted controls, edits a clone, hidden test, stamps threads and diffs;
checked predicted test run on the UNMODIFIED base related tests, one escalation catches what the tests passed
mechanically steps, a test
Opt-in, and costly: a sweep of four units (11,053 lines) ran about $90 at list price. Use it
when you will act on what the agents find. A single bug, a lookup, or a docs edit does not need it.
Method, numbers and caveats in full: docs/guides/AGENT_PIPELINE.md.
What was measured (four units, one run, 2026-10-06; list-price estimates)
| result | |
|---|---|
| Mapper as a saving | No. Opus tagging only the sonnet-pointed regions (31-50% of lines) cost 0.94-1.12x opus reading the whole unit: $0.60 vs $0.53 per confirmed finding. |
| Mapper as diversity | Yes. The mapped pipeline found 40 distinct verified clusters of 51, the smart-only one 34; 17 and 11 were unique to each. Union the two if recall matters. |
| Haiku as the action tier | 43 of 45 tasks passed the tagger's test with no regression; 35 of 45 after a sonnet read of the diff; 2 escalations to sonnet, both passed. |
| Tagger predictions | 8 of 8 predicted answers matched the human's, but the prediction was shown first and labelled Recommended, which can anchor the pick. |
| What the mechanical stages missed | 10 of 45 diffs flagged by a reviewer although the tests passed; 15 thread-level mistakes, including "all tests pass" reports the thread contradicts and new tests never run against the unfixed code. |
What ships with it
14 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- prompts/action.md 728 B
- prompts/cluster.md 719 B
- prompts/diff-review.md 1.3 KB
- prompts/forger.md 1.4 KB
- prompts/mapper.md 1.1 KB
- prompts/skeptic.md 1.5 KB
- prompts/tagger.md 3.0 KB
- README.md 881 B
- schemas/tag.schema.json 4.7 KB
- schemas/verdict.schema.json 1.3 KB
- scripts/check_tag_quotes.py 2.5 KB runs code
- scripts/run_action.py 8.7 KB runs code
- scripts/run_base_tests.py 4.7 KB runs code
- scripts/validate_pointers.py 2.7 KB runs code
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.
- 8d ago First seen · 100 lines · 83 tokens per session scan A 343f15667088
agent-pipeline is a skill published in the GitHub repository EmpiricaAI/empirica (252 stars, last pushed yesterday), licensed MIT. It adds 83 tokens to every session and 1,810 once invoked, about $0.0003 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 2026-10-08.
Other skills, from other repositories
pre-pr
Prepare a Honcho change for a pull request to plastic-labs/honcho. Invoke before opening a PR, when drafting a PR body, when asked if a branch is PR-ready, or when filling the pull request template. Checks the linked issue, required tests and docs, then writes Description / Proofs / Fixes.
risk-based-code-review
Use when reviewing or verifying software changes before completion, release, merge, deployment, or acceptance, especially when the appropriate test depth is unclear or security, data, migrations, external interfaces, infrastructure, or user-visible behavior may be affected.
quality
PALADIN quality gatekeeper. Two-stage system: Stage 1 checks spec compliance (pass/fail, blocks everything on failure). Stage 2 scores project health 0-100 across 6 tiers. Issues verdict: SHIP IT / SHIP WITH CAUTION / NOT READY / BLOCKED. Evidence Before Claims: every finding must cite the file, line, and what was…
security-audit
Security review and hardening workflow — root-cause analysis of vulnerabilities, authentication and authorization checks, least privilege, input handling, secret hygiene, and security regression tests. Use when reviewing code for security, fixing a vulnerability, hardening a feature, or handling auth, permissions…
dev-rules
Development methodology rules (Clean Architecture, TDD flow, 3-Strike rule, decision priority) that govern planning, implementation, and review. Load before designing, staging, implementing, or reviewing code changes.
agent-harness-fault-injection
Use when an agent workflow needs deterministic recovery evidence for sandbox, MCP/tool, worker, checkpoint, memory, or orchestration failures.