Last30days

Before reading anything else in this file, check whether you loaded SKILL.md from the one known stale-clone location: Claude Code's marketplaces directory.

Run this check:

CLAUDE_CACHE_LATEST=$(find "$HOME/.claude/plugins/cache/last30days-skill/last30days" -mindepth 1 -maxdepth 1 -type d 2>/dev/null | sort -V | tail -1)
# Two cache layouts ship in the wild — nested ({version}/skills/last30days/SKILL.md)
# and flat ({version}/SKILL.md). Resolve to whichever shape actually exists.
CLAUDE_CACHE_SKILL_MD=""
if [ -n "$CLAUDE_CACHE_LATEST" ]; then
  if [ -f "$CLAUDE_CACHE_LATEST/skills/last30days/SKILL.md" ]; then
    CLAUDE_CACHE_SKILL_MD="$CLAUDE_CACHE_LATEST/skills/last30days/SKILL.md"
  elif [ -f "$CLAUDE_CACHE_LATEST/SKILL.md" ]; then
    CLAUDE_CACHE_SKILL_MD="$CLAUDE_CACHE_LATEST/SKILL.md"
  fi
fi
echo "CLAUDE_CACHE_SKILL_MD=$CLAUDE_CACHE_SKILL_MD"

If the SKILL.md path you just Read contains /.claude/plugins/marketplaces/ AND $CLAUDE_CACHE_SKILL_MD is non-empty, STOP and re-read $CLAUDE_CACHE_SKILL_MD before proceeding. Otherwise the SKILL.md you have is fine — continue.

Why this specific check: ~/.claude/plugins/marketplaces/last30days-skill/ is a git clone Claude Code auto-restores to origin/main on session start. It can lag the versioned cache by one or more releases. Three 2026-04-22 test runs (Linear, Coinbase) loaded SKILL.md from marketplaces/, ran --help from the same stale path, did not see the --competitors flag that existed in the cache, and fell back to a manual comparison plan. Result: 2 of 3 windows never invoked the feature they were asked to test. STEP 0 defends against that one Claude Code-specific bug.

Other install paths are fine: ~/.codex/skills/, ~/.agents/skills/, an npx skills add install dir, or a repo checkout are all valid load points - the resolver in Step 1 picks them up. Do NOT abort or hop on those paths.

SKILL CONTRACT — READ BEFORE ANY TOOL CALL

You are inside the /last30days SKILL. This is a specific research tool with a 1400+ line instruction contract (the rest of this file) that defines EXACTLY how to produce the research output. It is not a generic "last 30 days of X" research prompt. Do NOT treat /last30days as a search keyword you can improvise against.

Named failure mode (2026-04-18 public v3.0.6 0/8 regression): on 8 consecutive public invocations, Opus 4.7 treated /last30days as a generic research keyword and improvised. Every single run violated LAW 2 (invented titles like "The headline", "Kanye West: the last 30 days"), LAW 4 (section headers like "Why he is everywhere this month", "1. gstack dominates", "The 'Homecoming' peak"), or both. One run (Matt Van Horn) skipped Step 0.5 / Step 0.55 entirely and ran the engine bare with zero resolution flags. Another (Garry Tan) leaked a trailing Sources: block despite LAW 1 reinforcement at four tiers. Two runs (Peter Steinberger, Kanye vs Kim) landed on a stale ~/.openclaw/skills/last30days/ engine copy via a self-written path-discovery loop.

How v3.0.7 fixes it: three structural anchors.

1. The MANDATORY first-line badge (🌐 last30days v{VERSION} · synced {YYYY-MM-DD}) at the top of every response is the LAW 2 / LAW 4 enforcement anchor. See "BADGE (MANDATORY, FIRST LINE OF OUTPUT)" in the synthesis section.
2. The SKILL_DIR substitution in the engine Bash calls uses the directory of the SKILL.md the model just Read — no resolver list, no precedence walk. Whichever install the harness loaded SKILL.md from is the install whose engine runs. Aligns spec-with-code and works for any harness without enumerating its install path.
3. This preface tells you plainly: do NOT improvise. Follow SKILL.md top to bottom.