/flow-next:audit — agent-native memory staleness review

Memory entries decay. A .flow/memory/bug/runtime-errors/ entry logged six months ago might reference a renamed file, a deleted function, or a codepath that no longer exists. Without periodic review, the store accumulates zombie entries and memory-scout surfaces outdated advice.

This skill IS the audit. The host agent (Claude Code / Codex / Droid) walks .flow/memory/, reads each entry, uses Read/Grep/Glob/git to verify references against the current codebase, applies engineering judgment, and decides per entry whether to Keep / Update / Consolidate / Replace / Delete. Optional autofix mode applies unambiguous actions and marks ambiguous as stale.

Decision entries (.flow/memory/knowledge/decisions/) and glossary terms (GLOSSARY.md files at the repo root and on the ancestor chain) are walked alongside the rest of memory. Decisions get a calibrated judging question — "does the constraint that motivated this choice still hold?" — and Replace becomes a two-step supersession (write successor, mark old decision_status: superseded, never git rm). Glossary terms are scanned for code usage; zero-hit terms get a <!-- stale: ... --> HTML comment via Edit tool (no flowctl glossary mark-stale exists), _Avoid_ aliases appearing in code surface as alias-creep findings.

There is no Python audit-engine, no codex/copilot subprocess dispatch, no deterministic scorer. The host agent is already an LLM and does the work directly. flowctl provides only thin persistence plumbing (memory mark-stale, memory mark-fresh, memory search --status) — landed by Task 2 of this spec.

Read workflow.md for the full phase-by-phase execution. Read phases.md for the 5-outcomes lookup with memory-schema-specific calibration.

Preamble

CRITICAL: flowctl is BUNDLED — NOT installed globally. which flowctl will fail (expected). Define once; subsequent blocks (here and in workflow.md) use $FLOWCTL:

FLOWCTL="${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/scripts/flowctl"
[ -x "$FLOWCTL" ] || FLOWCTL=".flow/bin/flowctl"

Inline skill (no context: fork) — AskUserQuestion must stay reachable across phases. Subagents can't call blocking question tools (Claude Code issues #12890, #34592). Phase 3 (Ask) and Phase 6 (Discoverability check) both require user choice in interactive mode. (sync-codex.sh rewrites this to a plain-text numbered prompt in the Codex mirror.)

Mode Detection

Parse $ARGUMENTS for the literal token mode:autofix. If present, strip it from the arguments — the remainder is the scope hint.

RAW_ARGS="$ARGUMENTS"
MODE="interactive"
if [[ "$RAW_ARGS" == *"mode:autofix"* ]]; then
  MODE="autofix"
  # Strip token, collapse whitespace, trim.
  SCOPE_HINT=$(printf "%s" "$RAW_ARGS" | sed 's/mode:autofix//' | tr -s ' ' | sed 's/^ //;s/ $//')
else
  SCOPE_HINT="$RAW_ARGS"
fi

Autofix mode rules

• No user questions. Never call the blocking-question tool.
• Process all entries in scope. No scope-narrowing question. If no scope hint was provided, process every categorized entry.
• Attempt all safe actions. Keep (no-op), Update (write tool), Consolidate (merge + git rm subsumed), auto-Delete (only when code AND problem domain both gone), Replace (only with sufficient evidence to write a trustworthy successor).
• Mark ambiguous as stale. When classification is genuinely ambiguous (Update vs Replace vs Consolidate vs Delete) or Replace evidence is insufficient, run flowctl memory mark-stale <id> --reason "..." instead of guessing. Stale-marking writes are atomic and round-trip safe.
• Conservative confidence. Borderline cases get marked stale; never deleted on autofix.
• Always print the full report. The report is the sole deliverable — there is no user to ask follow-ups.

Interaction Principles (interactive mode only)

In autofix mode, skip user questions entirely and apply the rules above.

In interactive mode, follow these principles:

• Ask one question at a time via AskUserQuestion (call ToolSearch with select:AskUserQuestion first if its schema isn't loaded). Fall back to numbered options in plain text only if the tool is unreachable or errors. Never silently skip the question.
• Prefer multiple choice when natural options exist.
• Lead with the recommended option and a one-sentence rationale.
• Do not ask the user to make decisions before evidence is gathered — Phase 1 investigates first, Phase 3 asks.
• Group obvious Keeps and obvious Updates together for batched confirmation. Present Consolidate / Replace / Delete one at a time.

The goal is automated maintenance with human oversight on judgment calls — not a question for every finding.

Forbidden

• Auditing legacy flat files (.flow/memory/pitfalls.md, conventions.md, decisions.md at the memory root). Skip with a warning that recommends /flow-next:memory-migrate first. Report includes the skipped count.
• Auditing under _audit/, _review/, or any other _* directory under .flow/memory/.
• Deleting silently. Delete is reserved for unambiguous cases (code gone AND problem domain gone). Default to Replace or Consolidate when there's still value to preserve.
• git rm on superseded decision entries. Decision history stays on disk. Replace for knowledge/decisions/ entries means write a new entry and mark the old decision_status: superseded with superseded_by: <new-id> — never delete the old file.
• Deleting glossary terms. When a term has zero code hits, mark stale via Edit-tool HTML comment. Removing the term entry is the operator's call, surfaced in the report.
• Inventing flowctl subcommands beyond what fn-34 task 2 ships (memory mark-stale, memory mark-fresh, memory search --status). fn-38 task 2 ships only glossary {add,list,read,remove} — there is no flowctl glossary mark-stale; use Edit tool. Use Write tool + git for moves and deletes.
• Mass-renaming code from a glossary alias-creep finding. The audit reports file:line locations and stops there; code rename is the operator's call.
• Auto-committing without user awareness in interactive mode. Phase 5 detects git context and asks. Autofix uses sensible defaults.
• Setting context: fork — blocking-question tools must stay reachable.
• Running parallel replacement subagents. Investigation subagents can run in parallel for 3+ independent entries; replacement subagents run sequentially to protect orchestrator context.

Pre-check: local setup version

Same pattern as /flow-next:plan and /flow-next:prospect — non-blocking notice when .flow/meta.json setup_version lags the plugin version:

if [[ -f .flow/meta.json ]]; then
  SETUP_VER=$(jq -r '.setup_version // empty' .flow/meta.json 2>/dev/null)
  PLUGIN_JSON="${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/.claude-plugin/plugin.json"
  PLUGIN_VER=$(jq -r '.version' "$PLUGIN_JSON" 2>/dev/null || echo "unknown")
  if [[ -n "$SETUP_VER" && "$PLUGIN_VER" != "unknown" && "$SETUP_VER" != "$PLUGIN_VER" ]]; then
    echo "Plugin updated to v${PLUGIN_VER}. Run /flow-next:setup to refresh local scripts (current: v${SETUP_VER})." >&2
  fi
fi

Workflow

Execute the phases in workflow.md in order:

0. Discover & Triage — walk .flow/memory/{bug,knowledge}/<category>/, group by module / category, count,