Chalk Labs
Need help deploying agent-conduct?
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/wtsi-hgi-agent-conduct
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/agent-conduct
What's inside
- Skill name
agent-conduct- Description
- Mandatory safety rules for all agents. Read before starting any work.
- License
- MIT
- Instructions
- 157 lines in SKILL.md · view on GitHub
Files (4)
- SKILL.md
- openai.yaml · agents/
- post-merge-cleanup.md · references/
- shared-resources.md · references/
SKILL.md (first lines)
# Agent Conduct These rules apply to ALL agents regardless of skill. ## Workspace Boundary - Write only inside the repository, a designated Git worktree of the same repository, or a harness-provided scratchpad/temp directory. Actual harness restrictions still apply; this rule grants no extra filesystem access. Redirecting command output to `/dev/null` is fine. - Before any file-writing command, confirm the target is within one of those locations. Create worktrees with normal Git commands, not clones or edits to `.git/` internals. - For PR branch work, the workflow owner creates one shared worktree before the first worker starts. Reuse it for implementor commits, reviewer reads, local gates, pushes, pr-resolver fixes, and rebases. Workers never create their own worktree for that branch; resume in the recorded worktree. - Create new PR branches from the freshly fetched base, from the main clone: ```bash git fetch origin && git worktree add --no-track -b <topic>-<8 hex> ../<repo>-<topic> origin/<base> ``` Use a short topic and eight random hex digits. Keep the worktree beside the main clone, never inside a clone. For an existing branch, reuse its worktree or attach that branch without `-b`. Dependent branches use the recorded SHA under [routing](../bugfix/references/incidental-issues.md#dependent-local-branches). - Reserve harness per-agent worktrees for throwaway experiments whose results are reported back, not retained as PR branches. - Give every worker the worktree's absolute path and require branch work only there. Reviewers leave it unchanged; mutants and experiments belong in a [scratch copy](#scratch-copies) or scratchpad. - Record the branch and worktree path in its branch-owned checklist's queue entry. In committed records, make the path relative to the main clone. - Remove a merged PR worktree through [post-merge cleanup](references/post-merge-cleanup.md), never when a worker finishes. Keep unmerged worktrees and branches until the user or caller authorizes deletion, then apply the Git Safety deletion checks. ### Scratch Copies A scratch copy is a temporary detached worktree in a scratchpad/temp directory: ```bash git worktree add --detach <scratchpad-or-tmp path> <sha> ``` Include the branch's topic or checklist ID in the scratch copy's path. Workers may create these for read-only checks or throwaway edits (mutants, bisects, base comparisons). Never commit in them. Before reporting back, apply the Git Safety deletion checks, then `git worktree remove --force <path>`. Only their throwaway files and edits are authorized for discard. ## Scratch Work - Use language-appropriate test temp dirs (`t.TempDir()`, `tmp_path`, etc.). - If a temp file is truly needed, prefer the harness-provided scratchpad directory; otherwise use `.tmp/agent/` in the repo and clean up. - Put every new source file in an existing package. A stray `.go` or `.py` outside one confuses tooling. ## Terminal Safety Avoid triggering VS Code modal confirmation prompts: - No interactive commands (`ssh`, `less`, `vi`). Use `git --no-pager`, etc. - No `sudo`. No broad `rm -rf`. - No port-listening processes without background mode. ## Git Safety - NEVER push to `master`, `main`, or `develop` branches, including normal and force pushes. Check the destination ref, not just the checked-out branch. - On the first authorized push, set the feature branch's own upstream with `git push -u <remote> <branch>`. - Keep the task's feature branch rebased on the latest remote `develop`, unless the caller or PR specifies another base. If remote `develop` is confirmed absent, resolve the configured remote's default branch instead. A failed fetch is a blocker, not evidence that a branch is absent. Retain the resolved base throughout the task. Deferred branches with an unmerged dependency follow the base transition and publication hold in [routing](../bugfix/references/incidental-issues.md#dependent-local-branches). Fetch before checking whether it is current. When that base has advanced, rebase onto it instead of merging it into the feature branch. Resolve conflicts and rerun the applicable quality gates on the rebased result. - After those gates pass, push the rebased feature branch with `--force-with-lease`. The user gives standing authorization for this; no separate request or confirmation is needed. Deferred branches held for recording or dependency merge remain local under [bugfix routing](../bugfix/references/incidental-issues.md). Verify the remote feature-branch tip before rebasing and use it as the explicit expected SHA in the lease. If the lease fails, inspect and reconcile the intervening work before retrying; never use plain `--force`. - Do not push other branches unless the user asks or the task is working on a PR and a push is needed to update the PR, rerun checks, or request review, or the push applies the standing rebase authorization above. - Do NOT modify `.git/` internals. - Use targeted `git add <file>` over `git add .`. - Before deleting a clone, worktree, or local branch, inspect the work that deletion puts at risk. For a clone, check all local branches with `git log --branches --not --remotes` and inspect detached HEADs too. Check at-risk uncommitted changes, untracked/ignored files, and stashes. Every at-risk commit or file must be verified preserved outside the deletion scope, verified integrated, or explicitly authorized for discard. For squash merges, verify the merged PR head and resulting integration diff; account for commits beyond that head too. Linked worktrees share branches and stashes; unrelated surviving work need not be cleaned. Report what was checked and how at-risk work was accounted for. Retain unaccounted-for work. These checks do not authorize deletion. - After the hosting service confirms a task PR merged, apply [post-merge cleanup](references/post-merge-cleanup.md). The user gives standing authorization to delete its verified local task branch, remove its clean disposable worktree, and prune stale remote-tracking refs without asking again. This does not authorize remote branch or clone deletion. ## General - Do NOT install system packages.
From wtsi-hgi/agentskills, MIT.
More from wtsi-hgi/agentskills
- 0.592score
implementation-principles
wtsi-hgi
Shared cross-language delivery workflow and guidance for implementing the simplest sufficient solution, maximizing reuse of existing code, fixing root causes…
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
bugfix
wtsi-hgi
Orchestrates standalone or caller-batched bug fixes via implementor and reviewer subagents using TDD.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
orchestrator
wtsi-hgi
Orchestrates implementation, review, and user-path verification of phase plans via subagents.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
final-response
wtsi-hgi
Remove filler, repetition, process narration, fake enthusiasm, and AI-sounding prose from every chat message, and end each one with a "Your next action" line.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
go-conventions
wtsi-hgi
Shared conventions for Go projects.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
go-reviewer
wtsi-hgi
Review Go implementations against spec acceptance tests.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
deliver
wtsi-hgi
Only when explicitly invoked, take the requested change and every resulting branch, including incidental fixes, through PR review and verified user merges.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
phase-creator
wtsi-hgi
Creates phase plan documents from a spec.md Implementation Order.
SkillAFreshness: A★ 1Claude Code +9 - 0.552score
go-test-strength
wtsi-hgi
Prove a Go package's tests catch defects, by injecting faults one at a time and requiring the suite to fail.
SkillBFreshness: B★ 1Claude Code +9 - 0.547score
code-smells
wtsi-hgi
Named code smell baseline for reviewing changed code, from Fowler's Refactoring chapter 3.
SkillBFreshness: B★ 1Claude Code +9 - 0.547score
frontend-design
wtsi-hgi
Create distinctive, production-grade frontend interfaces that avoid generic AI aesthetics.
SkillBFreshness: B★ 1Claude Code +9 - 0.544score
nextjs-fastapi-implementor
wtsi-hgi
Full-stack TDD implementation for Next.js 16 + FastAPI projects.
SkillBFreshness: B★ 1Claude Code +9
Signals
- Stars
- 1
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Oct 4, 2026 (26 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/wtsi-hgi/agentskills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from wtsi-hgi
- 0.592score
implementation-principles
wtsi-hgi
Shared cross-language delivery workflow and guidance for implementing the simplest sufficient solution, maximizing reuse of existing code, fixing root causes…
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
bugfix
wtsi-hgi
Orchestrates standalone or caller-batched bug fixes via implementor and reviewer subagents using TDD.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
final-response
wtsi-hgi
Remove filler, repetition, process narration, fake enthusiasm, and AI-sounding prose from every chat message, and end each one with a "Your next action" line.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
orchestrator
wtsi-hgi
Orchestrates implementation, review, and user-path verification of phase plans via subagents.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
go-reviewer
wtsi-hgi
Review Go implementations against spec acceptance tests.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
deliver
wtsi-hgi
Only when explicitly invoked, take the requested change and every resulting branch, including incidental fixes, through PR review and verified user merges.
SkillAFreshness: A★ 1Claude 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 agent-conduct 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
agent-conduct skill. Agent Mart. https://agentmart.dev/skills/wtsi-hgi-agent-conduct (last updated 2026-10-24)