agentleFS
MostAshrafMostAshraf/ai-sdlc-harness/skills/workspace-config/SKILL.md

workspace-config

Skill20 starsChanged 3 months agoMITView on GitHub

Change one config section (overrides / provider / language / repos) on an already-bootstrapped workspace, without re-running the full /init-workspace interview. USER-ENTRY — invoke only when the user explicitly runs /workspace-config; never autonomously, never from a subagent (guard-enforced).

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 0 are distinctive, and 5 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.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every command below is ${CLAUDE_PLUGIN_ROOT}/bin/harness <verb> … — run it yourself via Bash.

    line 13
  2. Very specific:

    Not this skill's job: registering a brand-new repo (that's /add-repo — discovery, duplicate-name/path refusal, permission-allowlist refresh.

    line 17
  3. Very specific:

    Resolve the plugin root: if not already set (native Qwen first run — each Bash call is a fresh subprocess, so the probe prints the path for …

    line 24
  4. Very specific:

    Read .claude/context/overrides.yaml.

    line 42
  5. Very specific:

    If the init-section result carries a notice key, relay its text to the user verbatim.

    line 76

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill 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.

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

workspace-config

Every command below is ${CLAUDE_PLUGIN_ROOT}/bin/harness <verb> … — run it yourself via Bash. Never ask the user to type a harness command; the user only answers the questions below.

Not this skill's job: registering a brand-new repo (that's /add-repo — discovery, duplicate-name/path refusal, permission-allowlist refresh; this skill only changes an already-registered repo's path/name/test_cmd), and first-time setup (/init-workspace's interview).

0 · Confirm the workspace is actually bootstrapped

Resolve the plugin root if not already set (native Qwen first run — each Bash call is a fresh subprocess, so the probe prints the path for textual substitution). Same resolution logic as /init-workspace step 0:

R="${CLAUDE_PLUGIN_ROOT:-}"; [ -z "$R" ] && {
  for p in "$HOME/.qwen/extensions/ai-sdlc-harness/bin/harness" \
           "$HOME/.qwen/extensions/ai-sdlc-harness/.qwen-extension-install.json"; do
    [ -e "$p" ] && { case "$p" in *install.json)
      R=$(python3 -c "import json,sys;print(json.load(open(sys.argv[1])).get('source',''))" "$p");;
      *) R=$(cd "$(dirname "$p")/.." && pwd);; esac
      [ -x "$R/bin/harness" ] && break; }; done; }
[ -x "$R/bin/harness" ] || { echo "ERROR: set CLAUDE_PLUGIN_ROOT to the dir with bin/harness" >&2; exit 1; }
echo "PLUGIN_ROOT=$R"

Use the printed path in place of ${CLAUDE_PLUGIN_ROOT} below.

Read .claude/context/overrides.yaml. Missing, or no bootstrap_completed key: /init-workspace never finished — stop and send the user there. Proceeding anyway reports "the change is live," and the user only learns otherwise later, at a confusing bootstrap incomplete refusal the next time /dev-workflow runs.

1 · Ask which section, and what's changing

  • overrides — any top-level key in config/defaults/*.yaml other than provider/repos/language (status_mapping, subagent_models, quick_mode, naming + change_types, review_rounds, stall, repo_map staleness N, review_policy, security.scan_cmd/ severity_order, …). Most single-setting changes land here.
  • provider — work-item/git provider + specifics (stories_dir, github_repo, github_project + github_project_owner (+ github_project_repo, which keeps board ids bare issue numbers), ado_org, ado_project, …).
  • language — per-repo test_cmd / coverage_cmd / quarantine, under language.repos.<name> (not the whole-workspace test_paths/ test_closure). Quarantine drops a known-failing spec unrelated to any run from every test + coverage command (reason + since required, init-verify gates the shape, each run flags the exclusions): {"language": {"repos": {"fe": {"test_cmd": "npx vitest run", "quarantine": {"exclude_template": "--exclude {test}", "tests": [{"test": "t/a.spec.ts", "reason": "fails on main", "since": "2026-07-22"}]}}}}} — language is a full REPLACE (§2): re-send every repo's existing keys.
  • repos — repointing or renaming an existing entry's path. A path may be a checkout root or a subtree of one (<checkout>/frontend) — that's how a monorepo's logical repos are registered, one entry per root.

2 · Read the current section before writing it

overrides deep-merges — a call only needs the keys actually changing: init-section --section overrides --json '{"quick_mode": {"loc_max": 50}}'. If the init-section result carries a notice key, relay its text to the user verbatim. Exception: a list-valued key (review_policy is the only shipped one) still replaces wholesale — dicts recurse, list items don't — so resend the whole list, including entries you're not changing, or one silently vanishes.

provider/language/repos are full-replace: the write IS the whole file. Read .claude/context/<section>.yaml first (Read tool, never a Bash one-liner), splice in the change, resend the complete set — a partial payload, or a forgotten self-nesting ({"repos": {...}} — overrides is the one flat, non-nested exception), silently drops every entry you didn't restate. For repos/language this is NOT caught by init-verify if it zeroes the whole map (step 4 covers what verify does check).

Known gap: overrides only adds/updates keys, never unsets — removing one needs a direct edit to overrides.yaml; say so rather than improvising.

3 · Confirm a test_cmd change by running it, and mind two risks

Confirm any test_cmd change by actually running it in that repo — same rule /add-repo//init-workspace follow, never accept it unconfirmed. Run it at the registered path, which may be a subtree of a checkout: test commands run with the registered path as their cwd, so a command confirmed from the enclosing checkout root proves nothing about that logical repo. Repointing an entry between a checkout root and a subtree of it is a legal repos write (init-verify accepts either), but it moves the command's cwd — re-confirm that repo's test_cmd in the same pass.

Known risks: (1) init-section --section repos has none of /add-repo's collision checks — a duplicate name (case-insensitive: collides on-disk on a case-insensitive filesystem) or a path under two names (misattributes test_cmd/scan_cmd) writes cleanly; check the full map yourself before sending it. Renaming also orphans language.repos.<old-name>, security.scan_cmd.<old-name>, and .claude/context/repo-map/<old-name>/ — carry them to the new name or tell the user to. (2) Avoid repointing a path or changing a test_cmd while a /dev-workflow run has in-flight tasks against that repo — resolution is by exact registered path, so a run mid-flight can start failing with a confusing "no test command" error.

4 · Re-verify if the section could affect verify or permissions

init-verify covers provider, repos (≥1 registered, each path inside a git work tree — a checkout root or any subtree of one, which is how a monorepo's logical repos register), and language (test_cmd runnability) — not overrides. A repo:<name> detail reading <path> (subtree of <checkout>) is a pass stating where the git tree actually is, not a failure.

  • provider only: re-run init-verify, fix any failure.
  • repos (a path) or language (a test_cmd binary): re-run init-verify, then init-finalize too — that's what refreshes the permission allowlist (Read on the repo path, Bash on the test_cmd binary) to the new value; skip it and the OLD path/binary stays allow-listed while the new one doesn't, so a background/mode:auto run hits an unprompted permission stall the first time it touches the changed repo.
  • overrides only: no re-verify/re-finalize needed.

5 · Tell the user what changed

State the section, the exact keys written, and (if re-verified/re-finalized) the result.

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.