Improve Goal

Enrich the user's draft goal into one clear, measurable, verifiable completion condition an autonomous session can work toward and a fresh evaluator can confirm from the transcript alone. The result must be ready to drop straight into Claude Code's /goal command.

HARD OUTPUT LIMIT — NON-NEGOTIABLE: the improved goal — the text inside the returned code block, including all whitespace — MUST be strictly under 4,000 characters (the goal pack's budget_chars, inherited from /goal's own input ceiling). Aim for a comfortable margin (roughly 3,500 characters) to absorb estimation error. Before returning, count the characters of the improved goal; if it is 4,000 characters or more, compress — remove redundancy, tighten wording, drop non-essential clauses — and re-count until it is strictly under. Compress; never truncate mid-sentence.

Pipeline

Run these four steps in order.

1. Load the rule pack

Read ${CLAUDE_PLUGIN_ROOT}/rules/goal-rules.json (equivalently ../../rules/goal-rules.json from this skill's base directory). Every enrichment decision traces to a rule in that pack via finding_refs, resolvable in ${CLAUDE_PLUGIN_ROOT}/rules/findings-index.json.

2. Confirm the artifact kind

Invoking this skill is the explicit classification: the draft is a goal. If it clearly reads as a prompt (instructions for a task, not an end state) or a loop (a recurring cadence or open-ended monitoring task), surface that mismatch as the first clarifying question. Never silently convert it. In default single-turn mode, honor the explicit invocation: enrich the draft as a goal anyway, with the mismatch question folded in as the first line of the returned artifact (naming the kind it appears to be and offering the matching sibling skill). In --interactive mode, ask the mismatch question first, before any other gap.

3. Detect gaps

Run the draft against every rule's gap_check. Produce one internal gap list of {rule_id, finding_refs, description} records — carrying every finding ref the rule cites — feeding both elicitation modes.

4. Elicit and synthesize

Default (single-turn): fold each gap into the returned goal as a concise clarifying question. Nothing invented, no placeholder tokens.

--interactive (only when the argument includes the flag): render the same gap list as AskUserQuestion calls — at most 4 questions per call, batched until every gap is resolved — then synthesize with the answers folded in. If the user declines or skips a question, fall back to folding that gap into the goal as a clarifying question — never invent an answer.

Synthesize by applying every applicable rule: one measurable end state (a single condition may bundle several checks joined with "and" — one end state, not a step-by-step plan), verifiable from the transcript alone (require the goal to print the evidence), a stated verify command, constraints on what must not change, a turn/time bound, and explicit partial credit for multi-component goals. Keep task-specific criteria distinct from team-wide definition-of-done.

Output format

Return ONLY the improved goal inside a code block, prefixed with /goal so it is ready to paste straight into Claude Code — the same convention improve-loop uses for /loop. The /goal prefix counts toward the 4,000-character budget. Nothing else in the response. Exceptions: when the user explicitly asks for the rationale or grounding behind changes, append the per-change citations (rule id → finding @id) after the code block; in --interactive mode, AskUserQuestion turns precede the single code block.

/goal [improved goal here]

Examples

Improved: "/goal All tests in test/auth pass (pytest test/auth -q shown exiting 0) and ruff check src/auth reports no violations, with both command outputs printed in the transcript. Constraints: change only files under src/auth; add no new dependencies; do not modify or delete any test; do not force-push. Stop after 15 turns and report the remaining failures if the suite cannot be made green." </example>

Improved: "/goal README.md has a Quickstart section with copy-pasteable install and run commands, and markdownlint README.md reports no errors (output shown in the transcript). Confirm scope first if 'improve' means more than adding Quickstart plus passing lint. Constraints: edit only README.md; change no code or other docs. Stop after 10 turns and report blockers if lint cannot pass." </example>

Critical rules

• Return ONLY the code block with the improved goal — no analysis, no explanations, no "Changes Made".
• The improved goal must be COMPLETE and READY TO USE immediately with /goal, including any clarifying questions needed to proceed.
• Prefix the returned text with /goal — the goal must be pasted straight into Claude Code, not hand-assembled.
• Express the goal as a single completion condition, not a task list or plan.
• NEVER use placeholders like [MODULE], [TEST], [THRESHOLD]; NEVER create templates. If information is missing, fold a concise clarifying question into the goal; do not invent details.
• STRICTLY under 4,000 characters (the text inside the code block, including all whitespace) — compress and re-count until it fits; never truncate mid-sentence.
• Preserve the original intent.
• When the user asks why a change was made, cite the specific rule id and its finding_refs — resolved via rules/findings-index.json — not a restated heuristic.