idk-now

IDK Now

For when the user doesn't know what to do next. Not a silent autopilot: survey → questions → a concrete, achievable next step that serves the project vision.

Unlike recon (tactical: git scope → skill from this pack), idk-now is directional: recover purpose, then pick the smallest useful move — which may be a DecisionNerd skill, DocSlime / ProductFeeling / Impeccable, or a vendored skill in this repo.

Operating contract

Shared by every DecisionNerd/dev-skills skill. Canonical copy: handbook/concepts/14-operating-contract.md.

• Define done first. Before using tools, write the finish line in one or two lines: acceptance criteria (existing issue AC, BDD scenarios, tests, or contract; otherwise propose them and where they should live) and the evidence that will prove them. Re-check it before reporting done. Never report done on unverified work.
• Requested scope sets the finish line. A question ("is it ready?", "why is it broken?", "what next?") ends with the answer and a Next: line naming the exact next invoke. An outcome request ("fix", "finish", "land", "#42") continues through the chain (diagnose → fix → test → check-readiness → merge-it) until the outcome or a real blocker. Do not end a turn with "Do you want me to…?" for in-scope, in-repo work.
• Stop only for real blockers. Stop and ask only when you cannot continue without the user, or before: deleting data or unmerged work, force-push or history rewrite, changing anything outside this repository (GitHub objects, deployments, live data, production or paid resources, external services), or leaving the requested scope, unless the user's request already named that exact action. Keep the harness's permission prompts for risky commands. Otherwise keep going and put status notes in the same message as the next action.
• Ask well, once. For a genuine question use the harness's structured question tool when it has one (Claude Code: AskUserQuestion; Codex: request_user_input when the current mode supports it; T3 Code: the question tool) with concrete options; otherwise plain text with numbered options. Treat the answer as settled; do not re-open earlier verdicts, plans, or answers unless asked.
• Fan out when work is parallel. Use subagents for independent reads (repo survey, evidence gathering, per-option research, per-area audits) and independent verification (a reviewer that did not write the change). Writes stay single-owner per path set and sequential. Brief every child with goal, done-when, constraints, must-not, and return shape; verify each result before consolidating. Use Claude Code's Workflow tool only when orchestrating many subagents truly needs it; it is expensive. T3 Code: its delegated subagents — with optional worktree isolation for independent verification — cover this without the Workflow tool.
• Pick the model tier per child; defer to routing config. If the harness or user config already routes subagents, follow it and do not pass a model. Otherwise pick a tier per child: mechanical search or inventory → fast/cheap; implementation and evidence gathering → mid; planning, review, adversarial verification → top. Tier names differ per harness (Claude Code: haiku / sonnet / opus or fable; T3 Code: inherit by default or pick from its live provider/model catalog). Set the choice with the harness's spawn model param or agent frontmatter model: (per-harness details: skills/agents/references/sub.md). Where the harness cannot choose, children inherit or the harness picks; say which in the status note.
• Keep a checklist on long runs. For more than about five steps or work that crosses skills, keep TASKS.md at the repo root and tick items as they finish. Do not commit it unless the repo already tracks one.
• Close every run with three headings. Blocked on me (the one genuine question or blocker, else "none"); Changed (files, commits, GitHub objects, deploys, else "nothing"); Found (evidence, verdict, and Next: <exact invoke>).

Commands

Optional free-text after the command is a hint (“shipping feels stuck”, “docs are a mess”).

Routing

1. No args / “idk” / “what now” → guide.
2. First word is a command → that command.
3. Clear “just list skills” → skills; “why does this project exist?” → vision.

Read-only by default. Do not implement, open PRs, or mutate GitHub until the user accepts a follow-up skill/action.

Flow (guide / default)

Follow references/flow.md. Summary:

1. Brief survey (read-only, keep it short)

Environment — cwd, OS hints if relevant, whether this is a worktree, active branch, dirty/clean.

Repo — remote, default branch, package/stack signals, open PR on branch if any.

Docs / vision signals — skim in order until you can state a candidate purpose (do not dump files):

• README.md, AGENTS.md, CLAUDE.md, PRODUCT.md, VISION.md, docs/ (esp. DocSlime-style PRODUCT / DESIGN / REQUIREMENTS / strategy)
• .productfeeling/, FEELING.md if present
• Landing/marketing copy or apps/web hero only if docs are empty

Git history — recent commits/themes (git log --oneline -15), ahead/behind, abandoned branches if obvious.

Skills present — see references/skill-universe.md.

Write a tight Survey block (bullet list, not an essay). If vision is unclear, say so explicitly. For a large repo (many packages, deep docs), fan out environment/repo/docs/git reads to parallel read-only subagents (fast tier); cap total file reads per the survey budget in references/flow.md.

2. Guided questions

Ask 3–5 questions max, prefer multiple-choice when possible. Use the harness's structured question tool (Claude Code: AskUserQuestion) with the options listed below; plain text with numbered options as fallback. Cover:

1. Horizon — today / this week / this milestone / vague “make it better”
2. Constraint — time, risk, “must not break prod”, solo vs team
3. Energy — ship / fix / clarify / design / clean / explore
4. Blocker feel — unknown next, too many options, broken thing, missing vision, waiting on people
5. Optional: user type they care about right now (if product-shaped)

Do not interrogate if answers are already in the user message — skip asked questions.

For quick: at most 2 questions, or zero if enough signal.

3. Goal → next steps

From survey + answers:

1. State Project vision (1–2 sentences) — inferred or confirmed; flag confidence low/med/high.
2. State Achievable goal for this session/week — small enough to finish; clearly advances the vision (not busywork).
3. Rank Next steps (primary + 2 alternates), each with:

• exact skill invoke (issues create …, productfeeling audit …, $impeccable polish, docslime-fill, recon issue #n, …)
• one-line why it serves the goal/vision

1. Close with three headings: Blocked on me (none, or the one real question), Changed (nothing — read-only), Found: vision confidence, achievable goal. Next: <exact primary invoke>.

Skill universe (must consider)

When recommending, search across:

1. This pack (DecisionNerd/dev-skills) — agents, architect-it, check-readiness, diagnose-bug, document-it, fix-it, idk-now, issues, kiss, merge-it, milestones, observe-it, pulls, recon, refactor-it, repos, research-it, test-it, tidy-up, troubleshoot-app, …
2. DocSlime — docslime-init, docslime-fill, docslime-adr, docslime-kiss, docslime-install (product docs tree)
3. ProductFeeling — feeling/emotion-aware product design (productfeeling / /productfeeling)
4. Impeccable — frontend craft (impeccable craft/shape/audit/polish/…)
5. Vendored / local skills in the repo space — discover under .agents/skills/, .cursor/skills/, skills/, vendor/**/skills, packages/**/skills, or paths AGENTS.md names. Prefer repo-vendored copies when present.

Details and routing heuristics: references/skill-universe.md.

Prefer the smallest skill that unblocks the achievable goal. Do not recommend a swarm of skills.

vision

Survey docs + README + recent commits only; ask at most two questions if purpose is ambiguous; output vision statement + one achievable goal + one skill. Skip full skill inventory unless needed.

skills

Inventory only: DecisionNerd pack from this repo’s skills/*/SKILL.md (or installed list), plus DocSlime / ProductFeeling / Impeccable if installed or vendored, plus any other vendored skills found. Table: name | one-line job | installed?.

Related skills

• Already know the domain, need tactical next move → recon
• Clear issue to plan → recon issue
• Emergency agent thrash → agents slap (not this skill)

Grounding

This skill’s TTPs are grounded in current engineering baselines (DORA, GitHub Docs, Fowler/Beck, Google SRE & SWE book, OpenTelemetry, OWASP LLM / NIST AI RMF, Diátaxis — see handbook sources.md).

Under uncertainty, frame the goal before options (small batches / Beck clarity). Survey → vision-tied next skill beats thrashing tools.

Handbook card: handbook/practices/idk-now.md.