RonanCodes/skills/repo-mode

Repo Mode

Some of Ronan's repos are personal (his own GitHub, agent-native, GH issues are the backlog). Others are work (a separate Atlassian / Jira / Azure DevOps / employer GH project owns the backlog) and the agent flow needs to stay invisible — no GH issues created, everything in gitignored local files.

This skill is the canonical control point. The five swarm-pipeline skills (write-a-prd, slice-into-issues, ralph, planner-worker, agentic-e2e-flow) all defer to the resolution algorithm below before deciding their default --target / --source.

Modes

• personal — PRDs and slices publish as GitHub issues. ralph and planner-worker consume from gh issue list --label ready-for-agent. The agent-native Pocock pattern. Default for repos owned by RonanCodes or Simplicity-Labs.
• work — Nothing leaks to GitHub. PRDs go to .ralph/<name>/prd.md. Slices go to .ralph/<name>/issues/*.md. ralph --kanban and planner-worker consume from disk. .ralph/ is auto-added to .gitignore. The flow is fully invisible to the work GH/Jira/ADO project.

Resolution algorithm (the contract)

Any skill that needs to know the mode runs this exact sequence:

# 1. Per-repo override (highest precedence)
mode=""
[ -f .claude/repo-mode ] && mode="$(tr -d '[:space:]' < .claude/repo-mode)"

# 2. Global default
[ -z "$mode" ] && [ -f "$HOME/.claude/repo-mode" ] && mode="$(tr -d '[:space:]' < "$HOME/.claude/repo-mode")"

# 3. Validate
case "$mode" in personal|work) ;; *) mode="unset" ;; esac

echo "$mode"

If mode == unset, the calling skill MUST prompt the user once via AskUserQuestion, persist the answer (see "First-run prompt" below), then proceed. After persisting, the same algorithm returns the saved value on every subsequent call so the prompt fires exactly once per repo.

First-run prompt (what calling skills do when mode is unset)

1. Get the suggested default by inspecting the gh remote owner:

   owner="$(gh repo view --json owner -q .owner.login 2>/dev/null || true)"
   case "$owner" in
     RonanCodes|Simplicity-Labs) suggested=personal ;;
     "") suggested=personal ;;   # no gh remote at all → personal default (legacy local flow)
     *) suggested=work ;;
   esac
   

2. Ask via AskUserQuestion (header: "Repo mode"):

   "How should this repo be treated for agent-driven PRD / slice / Ralph flows?"

   Options (recommended option listed first):

   • Personal — publish PRDs and slices as GitHub issues (the agent-native Pocock pattern; this is the gh-remote-owner suggestion when owner ∈ {RonanCodes, Simplicity-Labs})
   • Work — keep everything local and gitignored (nothing leaks to the work GH/Jira/ADO project; PRDs go to .ralph/<name>/prd.md, slices to .ralph/<name>/issues/*.md)

3. Save the answer to .claude/repo-mode (per-repo, always):

   mkdir -p .claude
   printf '%s\n' "$answer" > .claude/repo-mode
   

4. Add .claude/repo-mode to .gitignore if not already present (so the choice doesn't leak into work repos):

   touch .gitignore
   grep -qxF '.claude/repo-mode' .gitignore || printf '\n# repo-mode (personal/work agent flow toggle)\n.claude/repo-mode\n' >> .gitignore
   

5. If mode == work, also ensure .ralph/ is gitignored:

   grep -qxF '.ralph/' .gitignore || printf '\n# Ralph local backlog (work-mode private)\n.ralph/\n' >> .gitignore
   

6. Ask one follow-up via AskUserQuestion (header: "Global default"):

   "Save <answer> as your global default for new repos too? (writes to ~/.claude/repo-mode)"

   • Yes — most of your repos are this kind, save it as the default
   • No — only set for this repo, prompt me again next time

   If yes:

   mkdir -p "$HOME/.claude"
   printf '%s\n' "$answer" > "$HOME/.claude/repo-mode"
   

7. Echo the resolution back to the user in one sentence: Repo mode set to <answer> (saved to .claude/repo-mode<, global default updated>).

Direct invocation

/ro:repo-mode                    # --check is implicit; print current mode + where it came from
/ro:repo-mode --check            # explicit check
/ro:repo-mode personal           # set per-repo to personal (writes .claude/repo-mode)
/ro:repo-mode work               # set per-repo to work (writes .claude/repo-mode + adds .ralph/ to .gitignore)
/ro:repo-mode personal --global  # set global default to personal (writes ~/.claude/repo-mode)
/ro:repo-mode work --global      # set global default to work
/ro:repo-mode --clear            # remove per-repo override; fall back to global default
/ro:repo-mode --clear --global   # remove global default

When the first positional arg is personal or work, write to .claude/repo-mode (or ~/.claude/repo-mode with --global). When work is set per-repo, also append .ralph/ to .gitignore if missing.

When --check (or no arg), run the resolution algorithm and print:

Repo mode: <personal|work|unset>
Source:    <per-repo .claude/repo-mode | global ~/.claude/repo-mode | unset>
GH owner:  <owner>  (suggested: <personal|work>)

If unset, also print: No mode set yet. Next swarm-pipeline skill (write-a-prd, slice-into-issues, ralph, planner-worker, agentic-e2e-flow) will prompt you once and persist.

Why this exists

Without it, every PRD interview ended with "publish as GH issue?" and the user had to remember per-repo to say no for work repos. One leaked issue on a work GH project is a real awkward conversation. This makes "stay invisible at work" a one-time decision per repo (or one global decision) instead of a per-invocation cognitive load.

Cross-references

• agent-native repo Pocock pattern → [skill-lab:agent-native-repo-pocock](obsidian://open?vault=llm-wiki-skill-lab&file=wiki%2Fpatterns%2Fagent-native-repo-pocock)
• Calling skills that defer to this: /ro:write-a-prd, /ro:slice-into-issues, /ro:ralph, /ro:planner-worker, /agentic-e2e-flow