Imported from cardinalhq/cardinal-agent-plugins (adapters/gemini/skills/optimize-toolkit/SKILL.md). Install upstream with npx skills add cardinalhq/cardinal-agent-plugins --skill optimize-toolkit. Copyright stays with the author.

Cardinal Optimize Toolkit

Use this skill when the user explicitly asks Gemini CLI to optimize their toolkit, mine their own session history for capability-fit recommendations, or review whether a recurring inline pattern should become a reusable sub-agent. It is not a skill to reach for opportunistically just because the conversation is adjacent to agents or tooling — see What not to do below for why.

A per-user, past-informed, future-effective optimizer — not a session-local one. It looks at your own last 30 days of sessions, picks the top few recurring inline-work patterns worth acting on, authors the artifact itself grounded in your actual toolkit, and — only on your explicit confirmation — writes the accepted artifact to this working tree. Anything written takes effect next session, not this one: Gemini CLI loads custom subagent definitions (.gemini/agents/*.md) at startup, the same as ~/.gemini/settings.json and the extension directory — there is no live-registration channel, so accepting a candidate here will not change what happens for the rest of this conversation. (/agents reload reloads the agent registry without a full restart if the user's Gemini CLI version supports it — mention it as an option, but the "takes effect next session" framing below is still the safe default to tell the user.)

This burns your own session tokens to run. The server (conductor's maestro, via the cardinal MCP server) provides evidence — clusters, model mix, toolkit adoption, tier pricing, priced-savings math. You (Gemini CLI) do the authorship from first principles: pick the recommendation kind by reasoning about the evidence, then write the .gemini/agents/<name>.md artifact grounded in this user's real capabilities, real tool signatures, real cluster labels. There is no server-side artifact template — the artifact is composed here, per invocation, and passed back to the ledger verbatim on mark.

Before you start

This skill orchestrates six outcomes__* tools served by the cardinal MCP server (see the cardinal-connect skill — same server, same consent). They are documented in conductor's docs/specs/optimize-toolkit-mcp-tools.md:

1. outcomes__my_turn_pattern
2. outcomes__my_toolkit_adoption
3. outcomes__cluster_spawns
4. outcomes__org_offered_tiers
5. outcomes__estimate_savings
6. outcomes__mark

outcomes__my_recent_spawns also exists on the same server — it's a raw spawn history for ad-hoc debugging, not part of this flow. Reach for it only if the user explicitly asks "what did I spawn recently?".

Check your available tools before doing anything else. If none of the outcomes__* tools appear in your tool list, say so plainly and stop:

"The optimize-toolkit tools aren't wired into this org's Cardinal MCP server yet — that's a known rollout gap, not a problem with your connection. Nothing to do here yet."

Do not attempt a substitute analysis (no reading transcripts, no grepping local logs, no falling back to "let me look at your repo instead"). MCP unreachable or empty candidates → one line, stop, no retries.

How you (Gemini CLI) should run this

Budget: usually 5–7 tool calls total, ≤10 in the worst case. You'll also read a handful of files from the working tree to ground kind picking and authorship — that's fine, keep it targeted.

1. Open — situational-awareness bundle (2 calls)

Call outcomes__my_turn_pattern and outcomes__my_toolkit_adoption, both with window: "30d". From the two responses, compose a short (≤6 line) opening paragraph before mentioning any candidate:

• State the evidence window explicitly ("based on your last 30 days of sessions") — never let the opening imply "this session."
• Name the caller's top 1–2 model-mix rows by cost from my_turn_pattern.models.
• Name the toolkit-adoption headline the first candidate will target, e.g. "231 Explore-shaped spawns / 32.6M subtok in the last 30 days" — from whichever my_toolkit_adoption surface has the biggest signal this window: .agents, .skills, .commands, .mcp_servers, or .tool_counts. Built-ins count on equal footing with namespaced agents here — see step 5.
• No usable evidence, stop before any ratio math. If my_toolkit_adoption.coverage.sessions_scanned == 0, or my_turn_pattern.turns_total is 0, there is nothing to compute a coverage ratio from — say "no optimization candidates right now, not enough session history yet" and stop.
• Coverage caveat. Use my_toolkit_adoption.coverage.sessions_with_tier_attribution / coverage.sessions_scanned as the enrichment-coverage proxy. When that ratio is under 0.5, prepend a one-line caveat: "evidence from N/M of your sessions in the last 30 days — coverage will climb as you keep working on a current plugin version." Also check my_turn_pattern.coverage.plugin_versions_seen — if it shows a stale version for recent turns, mention the plugin-version-drift possibility and suggest a Gemini CLI restart.

2. Discover — pull raw spawns (1 call)

Call outcomes__cluster_spawns with window: "30d", min_jaccard: 0.99, min_cluster_size: 1. Those thresholds effectively disable the server's built-in token-Jaccard clustering — each spawn returns as its own single- member "cluster", giving you the raw spawn population (label + session_id + tool_signature + tokens + model). You do the actual clustering client-side in the next step, because the server's token-Jaccard is too coarse for semantically-similar-but-token-diverse labels ("Trace polly cwd flow" and "Map maestro sites/org API" are both investigation-shape but share zero content tokens).

Note the coverage.with_description_share — if it's under 0.5, mention it in the caveat you already added in step 1 (older sessions from pre-enrichment plugin versions don't emit subagent_description; it's legacy drift, not a per-agent gap).

3. Reduce + semantically cluster (Bash + LLM pass)

Stage 3a — mechanical reduction (Bash). Save the cluster_spawns response to a temp file and pipe it through the reducer that ships alongside this SKILL.md. The reducer collapses N-identical labels, groups by first content-word (verb), sub-groups by dominant-tool signature, and collapses same-session bursts. It turns 100+ raw spawn records into ~15–30 sub-clusters — a small enough set for the semantic pass in 3b to reason over without paging in the whole raw population.

# Find the reducer that shipped with this skill. Portable across
# install locations.
REDUCER=$(find ~/.gemini/plugins ~/.gemini/skills ~/.gemini/extensions . \
  -name reduce_spawns.py -path '*optimize-toolkit*' 2>/dev/null | head -1)
[ -z "$REDUCER" ] && {{ echo "reduce_spawns.py not found"; exit 1; }}

# Feed cluster_spawns response as JSON on stdin; get verb-bucketed JSON out.
python3 "$REDUCER" < /tmp/spawns_raw.json > /tmp/spawns_reduced.json
cat /tmp/spawns_reduced.json

Output shape: { input_spawns_raw, zero_signal_spawns, sub_cluster_count, reduced_rows: [{verb, tool_shape, spawn_count, zero_signal_count, enriched_spawn_count, avg_tokens_per_enriched_spawn, tools_seen, unique_labels, top_labels[:8], tokens_total, session_count, burst_count, sample_top_by_tokens}] }. Rows come sorted by tokens_total descending. avg_tokens_per_enriched_spawn and tools_seen are what step 5's counterfactual-ratio and extract-vs-gap mechanics read directly.

Zero-signal rows are retained, not dropped. A row where zero_signal_count == spawn_count has no tokens/model on any member (pre-enrichment sessions or tracker entries) — tokens_total reads 0 and tool_shape reads <empty> for these, so don't use them for cost/tier math (adopt-savings, downgrade share, pin thresholds). But their top_labels are real, often bespoke, self-narrated Task descriptions — exactly the evidence the contrast-pair mechanic below needs. Don't filter these rows out of your own reasoning just because tokens_total is 0.

Stage 3b — semantic cluster (your reasoning, in-context). Read the reducer's output and group verb buckets into meta-clusters by intent, not by surface tokens. Verb buckets like code/silent-failure/silent/ test/comment/type/review/independent/fresh-eyes/general all fold into one Code Review meta-cluster; trace/research/investigate /find/explore/map/inventory/reconcile fold into one Research & Investigation meta-cluster. Expected meta-clusters for a typical engineer include: Code Review, Research & Investigation, Migration / Implementation, Testing & Validation, Planning / Organization. Not every meta-cluster surfaces for every user; only name the ones that clear a real token/recurrence floor from the data in front of you.

The semantic pass is your judgment — no server-side taxonomy. Use the reducer's rich per-bucket evidence (top_labels, session_count, tool_shape) to justify each grouping. A verb bucket that doesn't cleanly fit any meta-cluster stays as its own single-bucket meta-cluster; don't force-fit for symmetry.

Take the top K = 3 meta-clusters by tokens_total. If fewer than 3 clear meta-clusters exist, present fewer — do not pad.

Known coverage gap. cluster_spawns only covers Task-tool subagent spawns. Slash commands (visible in my_toolkit_adoption.commands) never show up here — a heavy user of a release-command won't see a "Releases" cluster no matter how the reducer runs. If the user asks about command patterns, say so plainly and point at my_toolkit_adoption.commands for the raw counts; don't fabricate a cluster from spawn data alone.

4. Ground yourself in the user's real toolkit (0–2 file reads per meta-cluster)

Before picking a kind for a meta-cluster, look at what the user actually has. my_toolkit_adoption lists capabilities by name and usage, but "a name in a usage map" is not the same as "a file on disk that will still be there next session." For each of the top-K meta-clusters:

• If my_toolkit_adoption shows an agent/skill whose name looks like it might cover the meta-cluster's dominant verb buckets + tool_shape, read the matching file under .gemini/agents/ (or the user-scoped equivalent) to confirm the fit before recommending adopt. Bad adopt recommendations happen when the name matches but the actual scope doesn't. If the candidate is a built-in with no file on disk, use its description from your own tool listing instead — expected, not a reason to skip it (step 5).
• If my_toolkit_adoption.mcp_servers shows an entry with err > 0.1 * n, that's evidence for a swap/pin flag on the MCP side — check ~/.gemini/settings.json for how that server is configured before saying anything actionable about it.
• If you're considering extract (mint new), grep under .gemini/agents/ for the meta-cluster's tool_shape — a similar-shape agent may already exist under a name your my_toolkit_adoption scan didn't surface.
• If you're considering pin/downgrade, locate the existing agent file so you can propose a minimal edit (change one model line) rather than overwriting the whole file.

Keep this bounded — one or two focused reads per meta-cluster, not a full-repo sweep. If the file isn't obviously there, that's information ("adopt target's name matched but the file isn't under .gemini/agents/ — treat as evidence to soften the adopt pitch, or reframe as gap").

5. Pick the kind, from first principles

Call outcomes__org_offered_tiers once — this tells you the org's actual cheap and reasoning model tiers. Never suggest a model the org isn't offered; if a tier is null, that door is closed for this org.

Drill into sub-clusters — meta-clusters are framing, not the recommendation unit. The meta-cluster tells you the organizing shape of an engineer's work (Research is heavy, Code Review is heavy); the actionable play lives one level deeper, at the sub-cluster level: the individual verb-bucket from the reducer with its specific top_labels, tool_shape, sample_top_by_tokens[].model, session_count, and burst_count. Pitching at the meta-cluster level alone produces truisms ("mechanical work should run on Haiku") — the non-obvious plays only appear when you compare a sub-cluster's shape against my_toolkit_adoption.

Strongest pattern (empirically): toolkit-consistency adopt, found via contrast pairs — across every capability surface, not just agents. A contrast pair sets a named capability's aggregate usage (from my_toolkit_adoption, the "routed" side) against a reducer sub-cluster whose labels credibly describe the same domain but never name that capability (the "bypassed" side). Do not look for the contrast inside a single spawn's tool_signature — an Agent/Skill call there reflects the parent turn's routing decision, not the child subagent's own trace, and tool_shape (top-2 tools by frequency) usually buries a lone Agent/Skill call under Bash/Read noise anyway. Cross-data-source contrast is the mechanic that actually surfaces evidence:

1. Build the routed side from every my_toolkit_adoption surface, not just agents. Keep any entry — agent, skill, command, or MCP server — clearing ≥10 spawns/invocations OR ≥1M subtok/tok. This explicitly includes built-ins with no file on disk — a built-in clearing 200+ spawns is a real routed path, not "just the generic fallback." Also check .skills (≥5 invocations OR ≥100k tok) and .commands (≥5 invocations) the same way.
2. For each kept capability, build a small domain-vocabulary set from its real behavior: Read the description: frontmatter for a file-backed agent/skill, or use the description string your own tool listing carries for a built-in with no file on disk.
3. Walk the reducer's reduced_rows — including rows where zero_signal_count == spawn_count (the reducer retains these; their top_labels are real evidence, often the strongest signal precisely because they're bespoke, self-narrated Task descriptions rather than toolkit boilerplate). Flag a row as a bypass candidate when its top_labels share ≥2 content words with a kept capability's domain vocabulary. Also check .tool_counts for recurring native-tool patterns echoing a sub-cluster's tool_shape.
4. Require ≥2 spawns, same session or sessions within the window. A burst under a few minutes (check at timestamps) is a stronger tell than a single spawn. Bonus confidence: the same verb-phrase recurring across sessions with only the trailing subject varying.
5. Compute the counterfactual ratio. Routed-side avg = the capability's own subtok / n (agents) or tok / n (skills/commands). Bypassed-side avg = the sub-cluster's avg_tokens_per_enriched_spawn (from the reducer, divided by non-zero-signal spawns only). Ratio = bypassed avg / routed avg. ≥3x is a real magnitude signal worth stating out loud even with no cohort/$ pricing.
6. State the caveat honestly: cluster_spawns never exposes subagent_type, so you cannot prove the bypassed spawns did NOT route through the named capability — only that their labels don't mention it. Pitch it as "these don't reference <capability> and look hand-rolled" rather than "you didn't use <capability>."

Once you have a pair, the play is adopt — the user has the capability, they're inconsistently reaching for it.

Actionable-only filter — governs every kind below, not just adopt. A recommendation only reaches the user if it maps to a concrete change they can accept or reject right now: a file to write, a settings.json/hooks.json edit, or a specific one-click settings-page toggle. "Type X next time," "consider Y," "wait for Z to recur," "worth investigating," and every other non-authorable prose recommendation gets dropped — not shown, not flagged, not noted as a caveat. Nothing. If every candidate you considered this run fails to clear this bar, the user's entire output is: "Based on your last 30 days, no optimization changes to make right now." — followed by one paragraph of situational awareness if relevant (top model, coverage caveat) and stop. See §7 for exactly how this renders (or doesn't).

adopt never ships as plain prose — escalate to a concrete artifact or drop it. "Stop spawning this inline pattern, reach for <capability> instead" is itself non-authorable prose under the filter above — it asks the user to remember something, not to accept/reject a change. A contrast pair earns the adopt classification honestly, but before you present anything you still have to convert it into an artifact:

• Slash-command alias (the escalation available in this adapter). Author a TOML file at .gemini/commands/<alias>.toml (project- scoped — it lands in the working tree, shows up in git diff) whose prompt field simply invokes the target capability, so the fix is "type /<alias>" rather than "remember to reach for <capability>." Mention /commands reload as the fast path to pick it up without a restart. This plugin's shipped hooks.json wiring (SessionStart, BeforeAgent, AfterModel, AfterTool, AfterAgent, PreCompress, SessionEnd) has no verified event confirmed to gate a subagent spawn before it proceeds with a block/ask decision, so a PreToolUse-style intercept — unlike the claude adapter — is not an option here; do not invent one.
• Neither fits honestly (no sensible alias name, the capability isn't something a single-command alias can front) → drop the recommendation entirely. Do not fall back to "reach for <capability> next time" prose as a consolation artifact.

Extract vs gap — where the bar actually sits. gap is not the default fallback when there's no adopt target. A single sub-cluster row clearing ≥2 spawns with a non-empty tool_shape is extract. If sibling rows share the same verb but split across tool_shape buckets purely from top-2 noise (check tools_seen — the full tool set per row; near-identical tools_seen across rows means the same underlying job), combine their spawn_count before applying the ≥2 bar. Only fall to gap when, after combining, the surviving tools_seen sets share fewer than 2 tool names, or no combination clears 2 spawns.

Tie-break — adopt beats downgrade when both fire. If a sub-cluster is both adopt-covered (an existing agent handles this shape) AND downgrade-shaped (running on reasoning tier with a mechanical tool_signature), pitch adopt alone. The existing agent's own model config is a separate concern; routing the work to the agent captures the primary win. Do not double-recommend.

For each of the top-K meta-clusters, reason about kind from the evidence in front of you (the semantic cluster label, its member verb buckets, its dominant tool shapes, its token magnitude, and the my_toolkit_adoption match you just grounded). There is no server-side kind gate — you pick, you justify. None of these kinds are agent-only — every kind below can target an agent, a skill, or an MCP server:

• adopt — cluster's tool_signature and label overlap an existing capability you saw in step 3 (agent or skill). Recommend the user reach for the existing thing consistently — but see the escalation rule above this list: this only ships as a .gemini/commands/ alias, never as plain "reach for X next time" prose.
• swap — cluster overlaps an existing capability, but that existing one is wrong shape/model — or, for an MCP server, the wrong integration for the job. Recommend replacing it. MCP-server candidates surface from .mcp_servers: an entry with err > 0.1 * n is worth flagging, but a swap needs an alternative to point at — absent cross-org data, you usually can't name one. When you can't, don't force a swap — and don't narrate the error rate either; with no alternative to point at there's no authorable change, so this candidate is dropped per the actionable-only filter (see §7).
• pin — cluster runs on a mix of tiers with the org's cheap tier already carrying meaningful share (say, ≥30% of the cluster's subagent_model occurrences) and no evidence the reasoning tier is load-bearing. Recommend pinning the existing capability.
• downgrade — cluster runs predominantly on the reasoning tier (say, ≥50% share) but the tool_signature looks mechanical (jaccard_within ≥ 0.6, small tool set). Recommend re-tiering down.
• extract — recurring inline pattern with no existing named capability to point at. Mint a new sub-agent. This is the one kind where you author a full new .gemini/agents/<name>.md file from scratch.
• consolidate — two clusters look like the same underlying job. No new file; present conversationally.
• gap (not presented to the user, no artifact) — cluster is real recurring work, but you cannot pick a kind honestly even after the "Extract vs gap" combining step above. No artifact, no confirmation question, no estimate_savings call, and — under the actionable-only filter — no user-facing mention either; see §7. Not every capability surface produces a play from a given window — that's a quiet window, not a mechanic failure.

Thresholds above are rules of thumb, not gates. Adjust when the cluster's specifics clearly warrant.

6. Estimate savings (1 call per non-gap candidate)

For each candidate that isn't gap, call outcomes__estimate_savings with the cluster's total_tokens, per-component tokens if you can derive them, current_model, and target_tier. Read estimate.assumptions.placeholder_output_ratio / placeholder_cache_ratio and estimate.estimate on every response — see Placeholder savings, honestly below.

7. Author + Present (no tool call; you write the artifact)

If a recommendation has no authorable artifact, it does not appear in the output — not as a footnote, not as "worth investigating," not as a caveat tacked onto a real one. Silence is the correct output for that candidate. If, after applying §5's escalation rule to every adopt and dropping every gap and no-alternative swap, zero candidates survive, the entire response is the evidence-window line plus: "No optimization changes to make right now." Nothing else. That is a complete, good report — do not pad it with observations to avoid looking empty.

Per kind:

extract — new file at .gemini/agents/<name>.md. You author:

• name: kebab-case, ^[a-z][a-z0-9-]*$, derived from the cluster's representative_label. If a same-name file already exists, disambiguate rather than overwrite.
• description: one sentence stating what work this handles and when Gemini CLI should delegate to it. Ground it in the cluster's actual observed work — not a template.
• tools: derived from the cluster's tool_signature. If the tool_signature is empty, omit the tools: line.
• model: the target_model_id from org_offered_tiers.
• Body: a short (≤10 line) system prompt describing the role, grounded in this cluster's specifics.

pin/downgrade — edit an existing file at .gemini/agents/<name>.md. The meaningful change is one frontmatter line: model: <target_model_id>. Author the dry-run as a one-field edit.

adopt — always a .gemini/commands/ alias, per §5's escalation rule, or not presented at all. Author the full alias-command TOML file content. Lead with the counterfactual ratio from step 5 as the evidence for why the alias is worth writing, then the confirmation question — same as any other file-writing kind, no special "behavioral, no confirmation needed" carve-out anymore. Never offer a settings-entry or other prose-nudge note as a substitute for the alias — not on request. If the alias escalation from §5 is not honestly authorable, this candidate is dropped, full stop — no signal-only paragraph, no gap consolation mention.

swap — edit or replace an existing file (.gemini/agents/*.md or ~/.gemini/settings.json for an MCP-server swap). Author the dry-run as the specific edit you'd propose.

consolidate — no automated file work. Present the two-candidate overlap and ask the user which capability should absorb the other.

No clean capability-level intervention exists → drop it, don't narrate it. Some patterns don't map onto .gemini/agents/*.md, .gemini/commands/*.toml, or ~/.gemini/settings.json at all — an MCP-server error-rate flag with no alternative to name is the clearest example. v6 gave this a "say so plainly" sentence; under the actionable-only filter that sentence is itself the banned non-authorable prose. Say nothing to the user about this candidate. (You may still call mark with status: "presented" for the ledger's sake — the ledger is not user-facing.) Never substitute a settings-note nudge instead.

gap — no artifact, and not presented. Do not state the pattern to the user — that framing is exactly the "worth investigating" prose the actionable-only filter bans. Silently call mark with status: "presented" and proposed_kind: "gap" for the ledger's sake, and move on. A run where every candidate lands on gap is a run with nothing to show the user — see the "zero candidates survive" output above.

Present the recommendation like a colleague on Slack, not a monitoring dashboard. The mechanic below (§5's five-step algorithm, my_toolkit_adoption cross-reference, sub-cluster reduction) is scaffolding — it stays in your head, not on the user's screen. What the user sees should read like "someone smart looked at my week and noticed X."

Ban list — never appear in user-facing prose. Do not use these internal-mechanic terms in the pitch:

• sub-cluster, verb-bucket, meta-cluster
• routed side, bypassed side, contrast pair
• cluster_spawns, my_toolkit_adoption, subtok, zero_signal_count, tool_shape, tools_seen
• signal-only, <X>-shaped signal
• Kind names as headings or prefixes ("Candidate 1 — adopt / agent: Explore" is the mechanic's structure, not the user's mental model)
• Tabular metric dumps as the lead (small tables INSIDE prose are fine when they carry evidence)
• Anything of shape "worth investigating"
• Anything of shape "wait for X to recur"
• Anything of shape "consider adopting Y" (or any other "consider doing Y" phrasing with no artifact attached)
• Coverage caveats presented as their own standalone paragraph or section — they only ever appear inline, as a one-line prefix to a real recommendation
• Plugin-version drift as its own bullet or paragraph, unless it is itself the recommendation — "your plugin version looks stale; restart Gemini CLI" is a fine actionable one-liner; a paragraph musing about drift with no restart instruction is not

Voice. Short paragraphs. Concrete session dates and IDs where they add credibility. Ratios, not percentages, when the point is magnitude ("17× cheaper per spawn" reads sharper than "94% cheaper"). The recommendation goes FIRST; the evidence justifies it, doesn't lead it.

Example — bad framing (dashboard):

Candidate 1 — adopt Explore for code-locating spawns

Routed side (my_toolkit_adoption.agents.Explore): 14 spawns, 1.75M subtok → ~125k subtok / spawn. Description: "fast read-only search agent for locating code..." Bypassed side (Research & Investigation sub-clusters whose labels + Bash+Read shape read as pure code-locating, not routed through Explore): [table of metrics]

Example — bad framing (behavioral nudge, no artifact — banned under the actionable-only filter):

Past week, one thing worth changing:

You reached for generic subagents to do code-exploration work 9 times, burning ~19M tokens. Two clear giveaways:

• Jul 15, session 9f21ee76: "Trace polly sub-agent cwd flow" then "Find sub-agent detection signals" — back-to-back, both on opus-4-7, 8.4M combined
• Jul 15, session 2bbbd04f: three "Read telemetry" spawns fired in 4 minutes at 09:29 (sonnet-5, 2.3M combined) — a hand-rolled fan-out

Your Explore agent is designed for exactly this shape of work (locating code, finding files, grepping symbols). You used it 14 times last week at ~125k tokens each — 17× cheaper per spawn than the generic ones.

Next time you're about to spawn a Task starting with "Trace X," "Find X in code," "Read X telemetry," "Survey path," or "Inventory constants for X" — reach for Explore.

Can't quote dollars — your organization only prices Haiku and Sonnet 4.6, and you mostly run on opus. The 17× is a magnitude signal, not a receipt.

The evidence and the ratio are exactly right — that's real signal. But the closing ask ("reach for Explore next time") is a "type X next time" pitch with no file, no diff, no confirmation. Under v7 this either escalates to an artifact (below) or gets dropped; it never ships as-is.

Example — good framing (escalated to a command alias):

Past week, one thing worth changing — and I've written the fix:

Nine times in the last 7 days you spawned generic subagents to do code-exploration work Explore is built for — same pattern flagged two weeks ago too, so a "reach for Explore" reminder clearly isn't sticking on its own:

• Jul 15, session 9f21ee76: "Trace polly sub-agent cwd flow" then "Find sub-agent detection signals" — back-to-back, both on opus-4-7, 8.4M combined
• Jul 15, session 2bbbd04f: three "Read <adapter> telemetry" spawns fired in 4 minutes at 09:29 (sonnet-5, 2.3M combined)

Explore handles this shape of work at ~125k tokens/spawn — 17× cheaper than the generic ones. Since the reminder alone hasn't worked, I've written a command alias so it's one keystroke instead of a thing to remember:

New file — .gemini/commands/explore-this.toml:

description = "Delegate a code-locating/investigation task to Explore instead of spawning a generic subagent."
prompt = "Use the Explore agent to investigate: {{args}}"

Write this file? yes/no. (Run /commands reload afterward to use it without restarting.)

Present order per recommendation:

• If a recommendation has no authorable artifact → do not present it. No "signal-only" section, no "two things I noticed but won't recommend action on yet," no coverage-caveat rambles. Silence is the right output when there's nothing to do.
• If zero recommendations survive the actionable-only gate, the entire output is exactly: a one-line evidence-window note ("Based on the past N days (M sessions, K turns)"), then "No optimization changes to make right now." No further sections. That IS a good report.
• When there ARE recommendations, present each with an artifact/edit the user says yes/no to:
  1. Lead with what to change. One-sentence recommendation, in the user's own vocabulary (agent/command names, session labels).
  2. Show 1–3 concrete pieces of evidence — session IDs, dates, actual labels, tokens. Bursts and cross-session recurrence carry the most weight; solo spawns rarely justify a pitch.
  3. State the counterfactual as a ratio or a burst-collapse ("three spawns in four minutes → one call would have covered it"), never as a percentage. If dollars are unquotable, say so once and move on — don't apologize twice.
  4. Name the file(s) that will be written or edited — every surviving kind writes something now (adopt writes a command alias, per §5). If it's consolidate, ask which absorbs which.
  5. Confirmation question — every surviving candidate gets one; there is no more "behavioral, no confirmation needed" carve-out.
• Coverage caveats and plugin-version-drift notes only appear as a one-line PREFIX to a real recommendation (or as their own actionable one-liner, e.g. a restart instruction) — never as their own standalone section.

One candidate at a time — don't stack. If a file write needs a yes/no, wait for it before showing the next candidate.

Confirmation format when a file will be written. A plain question, target path visible ("write this to .gemini/agents/<name>.md? yes/no"), no ceremony.

8. Write (only on explicit confirmation)

Validate the target-file basename before anything else. Reject if the name (a) contains / or \, (b) contains .., or (c) doesn't match kebab-case ^[a-z][a-z0-9-]*$. On rejection, surface the value verbatim to the user with the specific reason and stop.

Only write after an explicit "yes"-shaped answer. No write on silence, hedging, or topic change. After writing, tell the user this takes effect next session (mention /agents reload as a faster option if the user's Gemini CLI version supports it). Consent, revert, and distribution are git.

9. Mark (1 call per candidate you presented)

Exactly one mark call per candidate. Pick from:

• status: "accepted" — confirmed and written.
• status: "dismissed" — explicit refusal only. Ask one short follow-up and forward the answer verbatim as reason (cap ~200 chars).
• status: "presented" — shown, no decision either way. Never auto-dismiss on non-confirmation.

Use action: { kind: "cluster", cluster_id, proposed_kind }. When you authored an artifact — extract/pin/downgrade/swap/ consolidate, or an adopt escalated to a .gemini/commands/ alias per §5 — pass agent_spec_md (your authored body, verbatim) and est_savings_low_usd / est_savings_high_usd (from estimate_savings) so the ledger row isn't lossy. For a dropped adopt (no escalation was authorable) and for gap, omit those fields — mark it regardless, even though nothing was shown to the user.

Mark is best-effort: if the call fails, don't error the conversation over it.

Failure handling for non-mark tools

mark is the one tool that follows the silent-log rule — every other tool is on the hard-stop rule. If any of my_turn_pattern, my_toolkit_adoption, cluster_spawns, org_offered_tiers, or estimate_savings returns 503, 400, or an empty result set where the flow depends on at least one row, surface the error verbatim, stop the flow, do not retry.

Placeholder savings, honestly

Every outcomes__estimate_savings response carries fields that exist specifically so you don't overstate a number:

• estimate: "no_cohort_catalog_only" means catalog pricing math only. Say so: "treat it as a ceiling, not a promise."
• assumptions.placeholder_output_ratio / placeholder_cache_ratio set to true mean typical-ratio fallback. Say "estimated within a wide band" rather than a bare point figure.
• current_cost_usd: null — dominant model not priced in this org's catalog. If the sub-cluster's dominant model (e.g., claude-opus-4-8) isn't in org_offered_tiers.all, estimate_savings returns current_cost_usd: null and savings_high/low_usd: 0. Do not quote a $ figure. Pitch the play on consistency or shape grounds ("mechanical tool-use pattern, better routed through ") — the savings surface just isn't usable in this org for this model.

What not to do

• Don't spawn a sub-agent to do independent investigation.
• Don't invent a cohort comparison when a tool response says there isn't one.
• Don't write anything without an explicit "yes" in this conversation.
• Don't auto-invoke yourself opportunistically. Gemini CLI has no hard invocation gate the way the claude adapter's disable-model-invocation: true provides — treat this prose rule as load-bearing.
• Don't paper over a bad pick with a plausibly-worded artifact. If you find yourself writing template-shaped prose, reclassify as gap, mark it, and drop it from the user-facing output (do not narrate the reclassification to the user).
• Don't reach for gap just because a sub-cluster's exact tool_shape row only has one spawn — check whether combining same-verb sibling rows by tools_seen overlap changes the count first.
• Don't offer a settings-entry or other prose-nudge note as a fallback artifact — for any kind, on request or not. The only honest fallback when no capability-level file applies is dropping the candidate silently — never narrate the absence of an artifact to the user.
• Don't exceed the ~10-call budget on outcomes__* tools.