configure - SKILL.md Agent Skill

name: configure description: One-time interactive configurator for the qship skill catalog. Run this immediately after installing the qship plugin (the SessionStart hook will remind you on first use). Walks through a short questionnaire (7 rounds, ~two dozen questions) plus a per-repo loop (1 repo or 100, the configurator doesn't care), renders the full skill set into your ~/.claude/skills/ tree, then verifies the install and required external plugins. Idempotent — re-runnable any time you want to change a value (added a service, new Jira project key, opted into a previously-skipped integration).

qship — interactive configurator

This skill is the bootstrap step for the qship plugin. The plugin ships templates; this skill collects the values that fill them in.

When to run

First time you install the qship plugin — the SessionStart hook prints a prompt directing you here.
Whenever you want to change a configured value — added a service, new Jira project key, opted into a previously-skipped integration (Alembic migrations, ngrok, Playwright). Re-running re-renders the whole catalog from the latest answers.

What it does, end-to-end

Loads previous answers from $CLAUDE_PLUGIN_ROOT/answers/last-run.json if present.
Asks the questionnaire (7 rounds, ~two dozen questions) plus a per-repo loop via the AskUserQuestion tool. Defaults pulled from $HOME, git config, $USER, and platform conventions wherever possible.
Writes the consolidated answers to $CLAUDE_PLUGIN_ROOT/answers/last-run.json.
Shells out to setup.sh --config <answers-file> to render templates via envsubst, write repos.json to the skill root, merge hooks into ~/.claude/settings.json, and touch the first-run marker.
Verifies the install via setup.sh --check and surfaces any missing external plugins to the user.
Prints a quickstart with example invocations.

Step-by-step

Step 1 — Load previous answers, if any

ANSWERS_FILE="$CLAUDE_PLUGIN_ROOT/answers/last-run.json"
mkdir -p "$(dirname "$ANSWERS_FILE")"
if [ -f "$ANSWERS_FILE" ]; then
  echo "[/qship:configure] Loaded previous answers from $ANSWERS_FILE."
fi

Use the values from $ANSWERS_FILE as the per-question default in the next step.

Step 2 — Auto-detect what you can

USER_HOME="${USER_HOME:-$HOME}"
SKILLS_ROOT="${SKILLS_ROOT:-$HOME/.claude/skills}"
LOCAL_DB_USER_DEFAULT="${LOCAL_DB_USER:-$USER}"
GH_HOST_DEFAULT="${GH_HOST:-github.com}"
DEFAULT_BRANCH_DEFAULT="${DEFAULT_BRANCH:-develop}"
PRIMARY_LANGUAGE_DEFAULT="${PRIMARY_LANGUAGE:-python}"
CODEBASE_ROOT_DEFAULT="${CODEBASE_ROOT:-$(git -C "$PWD" rev-parse --show-toplevel 2>/dev/null || echo "$HOME/work/your-monorepo")}"

Step 3 — Ask the questionnaire (6 rounds)

qship installs the full 21-skill pipeline every time — there are no skill-selection rounds. You only collect values (identity, repos, tracker, DB), never which skills.

Round 1 — Project identity + issue tracker (4 questions):

Company / project short slug — default acme.
Issue tracker — jira or none (default jira). jira uses the Atlassian MCP to fetch/create/transition tickets and read Confluence TRDs. none means no tracker MCP: you paste a ticket/TRD as text or pass a local markdown-file path, and the ticket-driven skills skip every Atlassian call. (linear/github are accepted but currently treated as none — no provider integration ships yet.) Sets tracker.tracker_type.
Ticket-key prefix — default PROJ. Used in ticket IDs / branch names / examples (e.g. PROJ, TASK, ABC). Stored as tracker.jira_project_key for placeholder back-compat; applies to any tracker.
GitHub host — default github.com. (You can fold GitHub host + org into this round; the configurator may ask them here or in a follow-up — keep total ≤4 per AskUserQuestion call.)

When tracker_type is not jira, skip the Atlassian cloudId question in the later rounds (it's Jira-only) — mirror what setup.sh does.

Round 2 — Codebase paths + GitHub org (4 questions):

GitHub org name — default your-org.
Absolute path to monorepo root — default $CODEBASE_ROOT_DEFAULT.
Top-level source directory inside each repo (e.g. src/, app/, packages/) — default src.
Default base branch — default $DEFAULT_BRANCH_DEFAULT.
State root for transient qship files — default /tmp/<slug>-qship.

Round 3 — Language + tooling (4 questions):

Primary language — python / typescript / go / mixed (default python).
Test command — language-specific default.
Lint command — language-specific default.
Format command — language-specific default.

Round 4 — Build the flexible repos[] list (per-repo loop — do this BEFORE the opt-in rounds):

Drop ALL assumptions about how the user's codebase is organised — single repo, 2-repo split, 12-service monorepo are all valid. Loop with AskUserQuestion until the user signals "done". For each repo capture:

Field	Description	Default
`name`	Directory name relative to `codebase_root` (or absolute path)	(required)
`kind`	Hint: monolith / service / library / gateway / frontend / worker / test-harness / other	`service`
`schema`	SQL schema this repo owns (null if N/A)	`null`
`package`	Python/JS package name for imports inside this repo	`null`
`has_migrations`	Set true if this repo has Alembic migrations — `qmigrationdevcheck` iterates only over flagged repos	`false`
`runs_locally`	Set true if `qspinuplocal` should start this repo's process for E2E	`false`
`is_primary`	The "main" repo for skills that need ONE default	`true` on first repo, else `false`
`port`	Default local dev port (integer or null)	`null`

Flow: (1) ask "How many repos?" (1 / 2-3 / 4+ / Other); (2) for each repo, grouped AskUserQuestion calls for name/kind/schema-package/flags+port; (3) "Add another repo?" yes/no. Single-repo users accept one entry and move on — everything MUST work with repos.length == 1.

Round 5 — Local DB (4 questions — used by qspinuplocal / qe2etest / qmigrationdevcheck):

Local Postgres role — default $LOCAL_DB_USER_DEFAULT.
Local Postgres host — default localhost.
Local Postgres port — default 5432.
Default local dev DB name — default dev_db.

Round 6 — Codex integration (1 question):

Wire skills into ~/.codex/skills/ as well as ~/.claude/skills/? — defaults y if codex is on PATH. (This is the only install flag; the 21-skill pipeline installs unconditionally.)

Step 4 — Compute derived values

COMPANY_SLUG_LOWER="$(echo "$COMPANY_SLUG" | tr '[:upper:]' '[:lower:]')"
COMPANY_SLUG_UPPER="$(echo "$COMPANY_SLUG" | tr '[:lower:]' '[:upper:]')"
CODEBASE_DIR_NAME="$(basename "$CODEBASE_ROOT")"
[ -z "${STATE_ROOT:-}" ] && STATE_ROOT="/tmp/${COMPANY_SLUG_LOWER}-qship"

# Env-var key names
ENV_SERVICE_URL_KEY="${ENV_SERVICE_URL_KEY:-SERVICE_URL}"
ENV_CLOUD_URL_KEY="${ENV_CLOUD_URL_KEY:-CLOUD_URL}"
ENV_OAUTH_CLIENT_ID_KEY="${ENV_OAUTH_CLIENT_ID_KEY:-OAUTH_CLIENT_ID}"
ENV_OAUTH_CLIENT_SECRET_KEY="${ENV_OAUTH_CLIENT_SECRET_KEY:-OAUTH_CLIENT_SECRET}"
DB_OWNER_ROLE="${DB_OWNER_ROLE:-app_admin}"

Step 5 — Write the answers file

Use the Write tool to emit answers/last-run.json matching config.example.json exactly. The repos field is the array you built in Round 4. Booleans under features are JSON true/false.

Step 6 — Run setup.sh non-interactively

bash "$CLAUDE_PLUGIN_ROOT/setup.sh" --config "$CLAUDE_PLUGIN_ROOT/answers/last-run.json"

This renders templates with the user's {{PRIMARY_REPO_NAME}} resolved, writes repos.json to $SKILLS_ROOT/qship/repos.json, merges hooks, and touches the first-run marker.

Step 7 — Verify the install + required external plugins (CRITICAL)

setup.sh already prints any MISSING external plugins. Run the self-check and surface the result to the user explicitly — this is the #1 cause of a first /qship run failing mid-pipeline:

bash "$CLAUDE_PLUGIN_ROOT/setup.sh" --check

If it reports missing plugins (superpowers, feature-dev, code-review, code-simplifier, pr-review-toolkit), DO NOT tell the user setup is complete. Show them the exact /plugin marketplace add … && /plugin install … lines and explain that qship's review/bug-hunt/simplify steps delegate to those plugins and will fail without them. Only declare success once --check exits 0 (or the user has acknowledged the missing-plugin list).

Step 8 — Print the quickstart

Show the user the example invocations + links to docs/CUSTOMIZING.md and docs/CODEX.md. Mention they can re-run /qship:configure any time to add a repo or change a value, and bash setup.sh --check any time to re-verify health.

How skills read the repos config at runtime

Multi-repo aware skills (qmigrationdevcheck, qshipmaster, qspinuplocal, qlocalclonedb) read $SKILLS_ROOT/qship/repos.json directly. Examples:

# Iterate every repo that has migrations
jq -r '.[] | select(.has_migrations==true) | .name' "$SKILLS_ROOT/qship/repos.json" \
  | while read repo; do
      cd "$CODEBASE_ROOT/$repo"
      alembic check
    done

# Resolve the primary repo
PRIMARY="$(jq -r '(.[] | select(.is_primary==true) | .name) // .[0].name' "$SKILLS_ROOT/qship/repos.json")"

# Get a repo by name
jq -r --arg n "$1" '.[] | select(.name==$n)' "$SKILLS_ROOT/qship/repos.json"

Single-repo aware skills (qcheck, qclean, qbug, qpr, etc.) operate on $(git rev-parse --show-toplevel) — no repos.json lookup needed.

Sanity checks

$CLAUDE_PLUGIN_ROOT must be set.
jq and envsubst must be on PATH.
$HOME/.claude/ must exist.
The repos[] array must have at least one entry.
Exactly ONE repo should be flagged is_primary: true (the configurator enforces this — defaults the first entry).

Notes for re-runs

Re-running this skill is safe:

setup.sh takes a timestamped backup of ~/.claude/settings.json before merging.
Re-rendering overwrites the existing ~/.claude/skills/<name>/<file> — any local edits the user made directly to a rendered skill will be lost.
The marker file gets re-touched.

Don't do these things

DO NOT skip AskUserQuestion rounds and silently re-use previous answers. The user invoked this skill because they want to change something — confirm each value.
DO NOT edit any file under templates/ directly — that's the maintainer-side template tree.
DO NOT assume a specific number of repos. Single-repo, 2-repo, 12-service: all valid.