agentleFS
adyusufadyusuf/claude-code-standards/skills/config/SKILL.md

config

Skill1 starsChanged 13 days agoMITView on GitHub

Shows or changes the adyusuf defaults that replace questions - working mode and the model for the Opus agents, whether unit tests and e2e specs are written, which of dev/test/prod are merged and pushed, and which tests run before a push. Also shows which agents the mode enables and the model each one runs on. Use it when the user says "/adyusuf:config", "change the defaults", "stop asking me", "don't push prod", "run e2e before test", "show the settings" or "which agents are active".

24 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use uninstall and resume.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • AnthropicSkill

    uninstall

    Uninstall a community skill that was installed via the hub. Confirms before deleting files, refuses to touch first-party plugin skills, and logs every action. Use when the user wants to fully remove a community skill ("uninstall [skill]", "remove this skill") rather than just disable it.

    Also in 3 projects that use this

  • genomewalkerSkill

    resume

    Resume a thread by loading its ~800-token context capsule

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 19 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When the user wants to change settings without naming them (no argument, "change the settings", "open the settings", "show the settings in …

    line 72
  2. Distinctive:

    Put the current value in each question or option ("now: yes"), because the tool cannot pre-select.

    line 85
  3. Very specific:

    Write all the answers with one apply --session <id> (validated all-or-nothing.

    line 100

“Common” is measured: kept by at least 6 in 10 of the 19 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

adyusuf configuration

Claude does not ask what a setting already answers (global rule #34). These settings are the answers. The user changes them here; everything else in this skill is mechanics.

Layers, lowest first: the plugin's defaults/adyusuf.json < the user's ~/.claude/adyusuf.json < the project's .claude/adyusuf.json (committed, like .claude/mode) < this session's own layer. A change typed as a command is written to the user layer unless the user says "for this project" (the project layer) or "for this session" (the session layer, --session <id>; the id is on the this session is <id> line of the start-of-session block).

The clickable questions and the Settings box's Session tab change the session layer only. A new session has a new id, finds no session file and starts from the saved defaults; nothing such a click did is kept. A press in the System tab writes the saved defaults (~/.claude/adyusuf.json), which every new session starts from. To keep a value, say so ("make it my default"): then set <key> <value> (user layer) or --project. A resumed session keeps its id, so its changes stay; /clear starts a new one.

The tool is ${CLAUDE_PLUGIN_ROOT}/scripts/adyusuf_config.py (python3):

The user saysRun
"show the settings"show (add --json for a machine view), then agents
no argument / "change the settings" / "open the settings"show, then the clickable questions below ("Changing settings with clicks")
"which agents are active"agents (--all includes the inactive ones): the roles that run and the model each runs on
"don't push prod" / "push prod: no"set push.prod no
"run the gates before pushing dev"set gates_before_push.dev yes
"run coverage before pushing dev"set coverage_before_push.dev yes
"wait for CI after pushing prod"set ci_after_push.prod yes
"run e2e on dev before deploying test"set e2e_before_deploy.test yes
"mode D, haiku"set mode D and set mode_model haiku
"write less" / "shorter replies" / "write more"set reply.length short / set reply.length long (normal between)
"turn qa off" / "developer on haiku"set agents.qa.enabled off / set agents.developer.model haiku
"all agents back to the mode"reset-agents (add --project for the project layer)
"for this project only"add --project
"for this session only"add --session <id> (works with set, unset, apply, reset-agents, and show/get/agents/context to read it)
"back to the default"unset <key> (add --project for the project layer)
several at onceapply <file or -> with a JSON object {"push.prod": false, ...}
"what can I change"schema
"update the plugin" / "get the new version"Run, in this order, with Bash: claude plugin marketplace update adyusuf-marketplace, then claude plugin update adyusuf@adyusuf-marketplace, then claude plugin list and report the version. Tell the user to run /reload-plugins (a plugin loads into a session only after that). The slash form /plugin marketplace update ... only opens the plugin page, which is why this goes through the shell. If the plugin is not installed at all: claude plugin install adyusuf@adyusuf-marketplace.

A value that is not valid is refused and nothing is written (exit 2); an unreadable file is never overwritten (exit 1) - say which file and let the user fix or delete it. After a change, print the changed line and say which layer it went to. Never edit the JSON files by hand when the tool can do it.

The settings, and who reads them

KeyValuesMeaning
modeA-EThe operating mode (modes/README.md). A per-session choice ($S/mode) or a project .claude/mode beats it.
mode_modelkeep, sonnet, haiku, fableWhat the Opus-defined agents run on (keep = as defined). Same override order as mode.
steps.enforceblock, remindSteps first (default block): every task is split into steps in the Status box before work; block refuses a code edit or a commit until they exist, remind only reminds. standards/00-working-method.md §1a.
reply.lengthshort, normal, longHow much Claude writes in a reply (default short); what each level holds: standards/00-working-method.md §5a. A red test, a skipped step and what the rules put in a reply (the long-run table, the #28 block) stay at every length.
agents.<role>.enabledauto, on, offOne of the 14 roles: auto follows the mode's set, on adds it, off removes it (mode A runs none).
agents.<role>.modeldefault, opus, sonnet, haiku, fableThe model that one agent runs on; default follows its definition and the mode's Opus override.
pod.connectyes/no"Connect to pod" (default no). A stored setting only: nothing in this plugin connects to anything or sends a message because of it (user decision 20/10/2026).
pod.message_formatshort, full"Message format" (default short). Stored only, like pod.connect.
write.unit_testsyes/noWrite unit tests when the code is done (#32).
write.e2eyes/noWrite e2e specs (optional to run, #33).
merge.dev merge.test merge.prodyes/noMerge finished code into that branch (#26, #34).
push.dev push.test push.prodyes/noPush that branch after its merge.
gates_before_push.<t>yes/noRun the gates (unit tests, coverage, secret scan, SAST: scripts/merge-gate.sh) on the merged result before pushing <t> (defaults: dev no, test yes, prod yes). Not run = recorded as Gate-Waived: gates on the merge commit (#34); coverage then runs alone (gate-core.sh <t> --coverage-only) if coverage_before_push.<t> is yes.
coverage_before_push.<t>yes/no"Run coverage before pushing": measure line coverage (≥ 80% per codebase, #29) before pushing <t> (defaults: dev no, test yes, prod yes). The gate reads it from the project's .claude/sdlc-settings.json; no = the step prints WAIVED and the merge commit carries Gate-Waived: coverage. A project keeps no coverage switch of its own.
ci_after_push.<t>yes/noAfter pushing <t>, wait for the project's CI to be green on the pushed commit; red stops the later targets (defaults: dev no, test yes, prod yes). Not waited for = Gate-Waived: ci on the merge commit. A project with no CI reports "no CI".
e2e_before_deploy.<t> (test, prod)yes/noBefore <t> is deployed, run the e2e suite on the environment below it (test: on dev, prod: on test); red blocks the promotion (defaults: both no). Applies only where the project has an e2e suite and that environment; otherwise the report says so.

Changing settings with clicks

When the user wants to change settings without naming them (no argument, "change the settings", "open the settings", "show the settings in the right-hand boxes"):

The rail boxes first. This plugin's mod (hooks/register.tsx, the boxes in hooks/boxes.ts) draws ONE pane beside the conversation with four bordered boxes, each folded by the arrow before its title, with nothing beside the title: Status (what runs now and next), Settings (a checkbox for every yes/no setting, a radio list for every choice), Agents (a checkbox per agent, a model radio list; folded at start) and Promotion (dev/test/prod, read-only; folded at start). A click in the Session tab runs adyusuf_config.py set --session <id> and changes this session only; one in the System tab changes the saved defaults. Open them with /adyusuf:panels, say that they are there and stop: the user changes the settings in the boxes, not in the chat. Never build a page or an Artifact for this. In the Claude desktop app the mod only loads when CLAUDE_CODE_PLUGIN_DIRS in ~/.claude/settings.json names the plugin's folder (SETUP.md §6); say so rather than falling back to a page.

Only when the boxes cannot be shown (the surface draws no panes, mods are off, the plugin is an old copy without them): run show, then ask with the AskUserQuestion tool, which draws clickable buttons, checkboxes and dropdowns in the chat. Put the current value in each question or option ("now: yes"), because the tool cannot pre-select.

  1. One question, multiSelect: true (checkboxes): "Which settings do you want to change?" with four options: Mode and model · Writing and pod (unit tests, e2e, pod.connect, pod.message_format) · Merge, push and gates · One agent.
  2. One AskUserQuestion call for each ticked group (at most four questions per call, two to four options each):
    • Mode and model: mode as buttons A / B / C / D (E through "Other"), then mode_model as buttons keep / sonnet / haiku / fable, then reply.length as buttons short / normal / long.
    • Writing and pod: write.unit_tests, write.e2e and pod.connect, each as yes / no buttons, then pod.message_format as buttons short / full (the pod settings are stored only; nothing acts on them).
    • Merge, push and gates: five checkbox questions (multiSelect: true, two calls): "Merge into", "Run gates before the push", "Push", "Run CI after pushing" (options dev / test / prod) and "Run e2e before deploying" (options on dev before deploying test, on test before deploying prod). A ticked box is yes, an unticked one no.
    • One agent: the role (qa, developer, architect, security, any other through "Other"), then enabled as buttons auto / on / off and model as buttons default / sonnet / haiku / opus (fable through "Other").
  3. Write all the answers with one apply --session <id> (validated all-or-nothing; the id is on the this session is <id> line), print show --session <id>, and say that it went to this session's layer. If the user asks to keep them, repeat the same apply without --session. Skip a question the user did not tick in step 1, and never ask again for what they answered.

When a change takes effect

At once, but never in the middle of a step. The files change immediately. A UserPromptSubmit hook and a PostToolUse hook (agent calls and shell commands, the task boundaries) run adyusuf_config.py notify: the first time a session sees different effective settings it receives a note naming what changed, and the note says to apply them from the next task and not to restart or split the step in progress. A note is sent once per change and per session; a session that never saw the settings stays silent.

What reads the settings

  • The SessionStart hook prints context: the effective values and the agents, so every session starts knowing them.
  • Rule #34 (CLAUDE.md): the merge/push/gate behaviour. A gate set to no is not run and recorded, never skipped silently; a red gate that was asked for still blocks. Force pushes, DROP, deploys and main stay out of reach of every setting (guard-destructive.sh, #20, #27).
  • adyusuf:working-mode falls back to mode / mode_model when no session or project choice exists.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.