behavior-repro
Use as step 1 when a complaint about an agent's behavior has to become evidence, before any cause is diagnosed or any artifact edited.
Skill by tdupu
Chalk Labs
Need help deploying behavior-repro?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/tdupu-behavior-repro
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/behavior-repro
What's inside
- Skill name
behavior-repro- Description
- Use as step 1 when a complaint about an agent's behavior has to become evidence, before any cause is diagnosed or any artifact edited. Not for code defects.
- License
- GPL-3.0
- Instructions
- 48 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Behavior repro A complaint is not evidence. Evidence is three written statements plus a fresh agent, holding nothing but those statements, doing the undesired thing. ## Write down three things - **Undesired** — what the agent did, decidable from the agent's output alone. "Called the change tested while its own report names no test it ran" is checkable. "Never actually ran the tests" is not: that is a claim about actions. Restate such a complaint as its output signature, or report it **unsettlable**. - **Desired** — what it should have done instead, stated so a reader of the output alone could tell the two apart. - **Situation** — the task, the inputs, and what the agent was working from. The complainant knows things the fresh agent will not. Those things go here or the repro will not run. ## Then confirm it reproduces Hand the situation to a fresh agent. Give it the task, never the complaint: a prompt that names the undesired behavior invites it, and you will have tested your own phrasing instead of the artifact. Behavior is stochastic: one clean run is not an acquittal, one bad run is not proof. Run it k times against the unedited artifact — default k=3, fixed here for the rest of the run — grading each run against the written undesired/desired pair and nothing else, and record the count. This count is the pre-edit floor, taken here, before anything has been changed; the later paired comparison in [[behavior-verify]] runs its own control arm, which must still clear this floor. **The floor is a majority of k.** Below the floor, stop here: before a cause is sought, before an edit is made. ## Terminal outcomes - **confirmed** — at or above the floor; what [[behavior-cause]] consumes. - **unreproducible** — never reproduced. - **underpowered** — reproduced below the floor. Give the counts and stop short of ruling on the complaint: from inside these runs a rare-but-real behavior and a bad repro look alike. - **unsettlable** — the situation cannot be described without naming the undesired behavior, or the only way to get the behavior is to ask for it. No repro, no repair: only **confirmed** continues; the rest are not a licence to edit prose and declare victory.
From tdupu/agent-skills-public, GPL-3.0.
More from tdupu/agent-skills-public
- 0.570score
assemble-recording-series
tdupu
Assemble one lecture or talk video from a folder of raw takes, whatever the capture tool wrote - flat screen-capture files, or per-session folders with the…
SkillAFreshness: AClaude Code +9 - 0.570score
check-paragraph
tdupu
Judge ONE paragraph of prose (LaTeX allowed) as its stated reader and return a fixed six-line verdict — topic sentence, one idea, reader can follow, claim…
SkillAFreshness: AClaude Code +9 - 0.565score
check-documentation-policy
tdupu
Audit mathcity documentation against POLICY-documentation.md.
SkillAFreshness: AClaude Code +9 - 0.565score
behavior-verify
tdupu
Use when an edit to a skill, an instruction file, or a prompt has to be shown to have changed the agent behavior it was meant to fix.
SkillAFreshness: AClaude Code +9 - 0.565score
check-citations
tdupu
Read-only citation audit for .tex files.
SkillAFreshness: AClaude Code +9 - 0.565score
check-grant-format
tdupu
Audit an NSF proposal folder against Grants/GRANT-FORMAT.md (page limits, fonts, margins, required headings, no URLs in the description, prior-support items…
SkillAFreshness: AClaude Code +9 - 0.565score
check-grant-style
tdupu
Audit an NSF Project Description and Summary against Grants/GRANT-STYLE.md (every project has a plan and names what is new; results in prose with citations…
SkillAFreshness: AClaude Code +9 - 0.565score
check-latex
tdupu
Produce the evidence block the human adjudicator needs to approve or reject a notes.tex (or any notes-tier .tex) change before push/merge.
SkillAFreshness: AClaude Code +9 - 0.565score
check-latex-hygiene
tdupu
Audit LaTeX beads, branches, or .tex diffs against the LaTeX Subdomain Policy (mathcity/subdomains/latex/POLICY.md, LX-rules).
SkillAFreshness: AClaude Code +9 - 0.565score
check-lmfdb-hygiene
tdupu
Audit an LMFDB-type bead, diff, experiment proposal, or live type against the LMFDB Subdomain Policy (mathcity/subdomains/lmfdb/POLICY.md) — labels valid…
SkillAFreshness: AClaude Code +9 - 0.564score
behavior-cause
tdupu
Use when a reproduced agent behavior has to be traced to the artifacts whose text produced it.
SkillAFreshness: AClaude Code +9 - 0.561score
break-up-proposition
tdupu
Break an existing proposition, theorem, or corollary into a labelled enumerate with separately signposted proofs while preserving its mathematics.
SkillAFreshness: AClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Oct 14, 2026 (15 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- GPL-3.0
- Repository
- github.com/tdupu/agent-skills-public
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from tdupu
- 0.570score
assemble-recording-series
tdupu
Assemble one lecture or talk video from a folder of raw takes, whatever the capture tool wrote - flat screen-capture files, or per-session folders with the…
SkillAFreshness: AClaude Code +9 - 0.570score
check-paragraph
tdupu
Judge ONE paragraph of prose (LaTeX allowed) as its stated reader and return a fixed six-line verdict — topic sentence, one idea, reader can follow, claim…
SkillAFreshness: AClaude Code +9 - 0.565score
check-citations
tdupu
Read-only citation audit for .tex files.
SkillAFreshness: AClaude Code +9 - 0.565score
behavior-verify
tdupu
Use when an edit to a skill, an instruction file, or a prompt has to be shown to have changed the agent behavior it was meant to fix.
SkillAFreshness: AClaude Code +9 - 0.565score
check-documentation-policy
tdupu
Audit mathcity documentation against POLICY-documentation.md.
SkillAFreshness: AClaude Code +9 - 0.565score
check-grant-format
tdupu
Audit an NSF proposal folder against Grants/GRANT-FORMAT.md (page limits, fonts, margins, required headings, no URLs in the description, prior-support items…
SkillAFreshness: AClaude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get behavior-repro running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
behavior-repro skill. Agent Mart. https://agentmart.dev/skills/tdupu-behavior-repro (last updated 2026-10-24)