until-hired / references
AGENTS.md2 starsChanged 3 days agoMITView on GitHub
Products change quickly. Treat these as starting points and confirm against what you can actually see and test.
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository13 other files it gives its agents.
- agent-setupSkill
- application-kitSkill
- apply-assistSkill
- candidate-intakeSkill
- growth-engineSkill
- inbox-syncSkill
- interview-coachSkill
- job-discoverySkill
- job-trackerSkill
- networkingSkill
- offer-coachSkill
- self-improveSkill
- until-hiredSkill
The file
We split this agents.md into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 31
If the product reads SKILL.md folders, copy the skills into its skills folder (python3 install.py --agent generic --target <folder>).
- Very specific: line 32
If it reads AGENTS.md, put the always-on block from agent-setup §6 there.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Notes per agent product
Products change quickly. Treat these as starting points and confirm against what you can actually see and test.
Claude Code (terminal, desktop or web)
- Install: as a plugin from this repository's marketplace (recommended: it adds the guard hook that blocks unsafe tool commands and asks before consent files change), or
python3 install.py --agent claude-code(skills and agents only, no hook). - Subagents: yes; the plugin ships
scout,applicant,inbox-keeper,researcherandreviewer. Parallel runs are supported. - Memory file:
CLAUDE.mdin the project folder. - Browser: a Playwright MCP server pointed at a dedicated browser profile, or the Claude in Chrome extension (which drives the candidate's own Chrome; agree which sites you may open).
- Notifications: hooks can run
bin/notify.pywhen Claude needs input or stops. Seenotifications.md. - Permissions: see
permissions.mdfor allow / ask / deny rules that fit until-hired. - Inbox: an email MCP server or connector, if the candidate has one.
Claude app (web, desktop, mobile)
- Install: upload the skill folders as custom skills, if your plan offers it, or paste
until-hired/SKILL.mdinto a project's instructions and attach the other skills as files. - Connectors (for example Gmail, Google Drive, Google Sheets) can serve as the inbox and a tracker mirror.
- Subagents, browser control, scheduled tasks and push notifications depend on the plan and the surface: check the tool list.
- Files persist only where the product gives you a persistent workspace. If it doesn't, keep the workspace in a connected folder on the candidate's computer. A synced drive works but sends private answers to that cloud too; say so and let the candidate choose.
Google Antigravity (agy CLI / IDE)
- Install:
agy plugin install https://github.com/Hrutin-Suriya-Kotni/until-hired, orpython3 install.py --agent antigravity(copiesskills/*to.agents/skills/andagents/*to.agents/agents/in the current folder). - Memory file:
AGENTS.mdin the workspace folder. - Subagents: yes. Don't list MCP tool names in a subagent's
tools:frontmatter (it can break the subagent's start); let it inherit MCP tools instead. - Browser: a Playwright MCP server in
.agents/mcp_config.json, attached over CDP to a separate Chrome profile. - Notifications: hooks (
~/.gemini/config/hooks.json) that runbin/notify.py(never a rawcurlwith the topic). Hooks may not fire for subagents' tool calls, so important checks live inside the tools, not only in hooks. - Guard: the Claude Code guard hook does not run here, so the safe-command rules are instructions only. Keep "ask before running commands" on for anything outside
<workspace>/bin/. - Approval prompts for ad-hoc commands slow long runs: prefer the workspace tools (
tracker.py,knowledge.py) over one-off scripts.
Other agents (Codex CLI, Gemini CLI, Cursor, GitHub Copilot, ...)
- Codex (experimental):
codex plugin marketplace add Hrutin-Suriya-Kotni/until-hired, orpython3 install.py --agent codex. - If the product reads
SKILL.mdfolders, copy the skills into its skills folder (python3 install.py --agent generic --target <folder>). - If it reads
AGENTS.md, put the always-on block fromagent-setup§6 there. - Otherwise paste
until-hired/SKILL.mdas the system or project instructions and keep the other skills as reference files. - Without subagents, use solo mode: same steps, one at a time, with the reviewer checklist.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

