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/ZIB-IOL/The-Agentic-ResearcherWrote 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/commands/zib-iol/the-agentic-researcher/retro)<a href="https://agentmods.dev/commands/zib-iol/the-agentic-researcher/retro"><img src="https://agentmods.dev/badge/commands/zib-iol/the-agentic-researcher/retro/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/commands/zib-iol/the-agentic-researcher/retro"><img src="https://agentmods.dev/badge/commands/zib-iol/the-agentic-researcher/retro.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 | $0.00010 | $0.01527 |
| Opus 5 | $0.00005 | $0.00763 |
| Sonnet 5 | $0.00002 | $0.00305 |
| Haiku 4.5 | $0.00001 | $0.00153 |
Grade A, and why
retro 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 10d 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 — 150 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are reflecting on the current research session to identify improvements for the agent's base instructions. This is a retrospective -- not about what went wrong, but about what can be made better for future research sessions.
User feedback (may be empty): $ARGUMENTS
Step 1: Gather Context
Detect which instruction file exists in the workspace:
- Check for:
CLAUDE.md,GEMINI.md,AGENTS.md(in that order) - Use the first one found as
$INSTRUCTION_FILE
Read the following files (skip any that don't exist):
/workspace/REVISION.md-- previous retrospective entries (if any)/workspace/report.tex-- experiment log, results, analysis quality/workspace/TODO.md-- open items, deferred work/workspace/$INSTRUCTION_FILE-- the instructions governing this session (especially Section 8: Project Instructions)- Run
git log --oneline -30-- see the commit history (style, frequency, quality) - Run
git diff --stat HEAD~5..HEAD 2>/dev/null || true-- recent change patterns
Backward compatibility: If /workspace/research_instructions.md exists (older format), read it as supplementary context for the original research goal.
Step 2: Commandment Compliance
Review compliance with each of the 10 Commandments (Section 1 of the instruction file). For each, assess: followed, partially followed, or violated, with evidence.
| # | Commandment | Status | Evidence |
|---|---|---|---|
| I | Never break a promise | ? | Did the agent follow through on stated intentions? |
| II | Never manipulate evaluation | ? | Were metrics/test sets/constraints kept unchanged? |
| III | Never fabricate citations | ? | Were bibliography entries verified? |
| IV | Complete all autonomous work | ? | Were tasks left incomplete? Did the agent wait unnecessarily? |
| V | Make it work before moving on | ? | Were methods discarded due to implementation bugs? |
| VI | One variable per experiment | ? | Were experiments properly isolated? |
| VII | Evaluate in tiers | ? | Was the three-tier strategy followed? |
| VIII | Bound your expectations | ? | Were theoretical bounds established before heuristics? |
| IX | Record everything | ? | Are results tables present? Are failures documented? |
| X | Verify before claiming | ? | Were verification scripts created for non-trivial math? |
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.
- 10d ago First seen · 150 lines · 10 tokens per session scan A 8c24a4f02c61
retro is a command published in the GitHub repository ZIB-IOL/The-Agentic-Researcher (95 stars, last pushed 2mo ago), licensed MIT. It adds 10 tokens to every session and 1,527 once invoked, about $0.0001 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 commands, from other repositories
skill-refresh
You are running inside a headless Claude Code session in the neo4j-skills repository. Goal: scan Neo4j release notes and deprecation pages, then update any SKILL.md and references/ files that are stale, outdated, or missing new features. Open no PRs or commits — output only file edits. The workflow will handle git and…
session-stop
Stop recording and save this agent session to the project ledger.
wtfp:resume-writing
Restore the latest portable handoff and route the researcher to the next safe action.
wtfp:discuss-section
Capture the researcher's intended purpose, essential content, voice, and boundaries before planning.
wtfp:new-paper
Interview the researcher and initialize a portable WTF-P project.
wtfp:verify-work
Test a drafted section against each plan requirement and record evidence one check at a time.