Use when explaining System V AMD64, ARM AAPCS, RISC-V psABI, stack frames, variadic calls, or FFI register rules. Not for the Rust FFI binding layer: use rust-ffi.
The English summary is being prepared. Showing the original description.
Use when asking "overkill?", "elegant?", "audit", or "taste-test this", or setting a taste register. Not for tasks that require source or remote-system changes.
The English summary is being prepared. Showing the original description.
See how to installReview the instructions the agent will receive before you install.
| Field | Bound contract |
|---|---|
| Trigger | User asks 'overkill?', 'elegant?', 'audit', or 'taste-test this', or asks to set a taste register |
| Authority | Read-only. No file, VCS, credential, paid, published, deployed, or remote mutation. |
| Side effect | Audit returns a verdict table and ranked fixes in chat; anchor mode persists the register best-effort; no files are written. |
| Done | A per-anchor verdict table with ranked top-3 fixes is returned, or the taste register is loaded in anchor mode. |
Auto-detect from the user's phrasing, with slash-arg override:
overkill?, elegant?, audit, taste-test this, judge this → audit mode.taste anchor, taste mode, taste register, or anticipates producing fresh work → anchor mode./taste audit or /taste anchor. Override always wins./taste anchor if drift is observed.Suspend the register temporarily for:
git push --force, rm -rf).Resume the register once the high-stakes section ends.
Side A: slop (centroid-AI default convergence): generic openers ("Sure!", "Of course"), hedge-stacks ("perhaps it might be"), validation phrases ("you're absolutely right"), AI-flat prose with no rhythm, default palettes, defensive nil-checks where impossible, 50/50 decision hedges that pick nothing.
Side B: overkill (decoration covering thin ideas): gradient stacks on every section, thesaurus-soup prose ("orchestrate the comprehensive synthesis of"), abstraction towers (4 layers where 1 suffices), complexity-flex masking absent conviction, ceremony that performs depth without delivering it.
| Domain | Side A (slop) | Side B (overkill) |
|---|---|---|
| Prose | "I think this might possibly help..." | "We orchestrate a paradigm shift across" |
| Code | try { x } catch { /* swallow */ } | Factory<Builder<Strategy<T>>> |
| Design | Default purple-blue gradient | Gradient on every section + glow + glass |
| Decision | "Both options have merit, so..." | 12-criterion weighted scoring matrix |
Eleven anchors apply across prose, code, design, and decisions:
| Anchor | Directive |
|---|---|
| Clarity | The artifact says what it means; reader does not have to decode. |
| Hierarchy | Important looks important; secondary supports. |
| Intent | Every choice is committed; nothing reads as "I let the default decide." |
| Coherence | Parts agree; tension only where deliberately staged. |
| Restraint | Default posture; compress before adding. |
| Generosity | Gives more than required at the right moment. |
| Honesty | No decoration covering missing depth; no slop covering missing POV. Care finishes the unseen. |
| One strong moment | Exactly one commitment carries the lift; the rest supports. |
| Invariants | The artifact names what must hold; preconditions, postconditions, and invariants are explicit. |
| Audience | The artifact enters through the reader's frame, not the author's. |
| Emptiness | The artifact leaves space the user completes; the least assertion. |
| Failure class | Response |
|---|---|
| Artifact too short or ambiguous to judge | Return "insufficient surface" with the specific missing element; do not fabricate verdicts. |
| Anchors produce conflicting fixes | Surface the tension; do not auto-pick. Tie-break is user-led. |
| Anchor mode drift after context compaction | Re-invoke /taste anchor; do not assume persistence. |
| User artifact is in a domain with no charter row | Apply the anchors directly; note the domain gap in the verdict table. |
No partial-result fabrication. If fewer than eleven anchors can be judged, return only the judged anchors and name the blocker.
Anchor | Verdict | Citation | Fix
-------------------|---------|-----------------------|----------------------------
Clarity | pass | |
Hierarchy | warn | Side A: AI-flat prose | Lead with the verdict line
Intent | fail | Side A: hedge-stack | Pick one; drop "might"
Coherence | pass | |
Restraint | warn | Side B: ceremony | Cut the framing paragraph
Generosity | pass | |
Honesty | pass | |
One-strong-moment | fail | Side A: 50/50 hedge | Commit to one direction
Invariants | pass | |
Audience | pass | |
Emptiness | pass | |
Top-3 fixes: 1. Pick one direction (Intent + One-strong-moment).
2. Lead with the verdict (Hierarchy).
3. Cut the framing paragraph (Restraint).
The register is loaded; subsequent responses hold the eleven anchors as imperatives. No table is emitted.
No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Use when explaining System V AMD64, ARM AAPCS, RISC-V psABI, stack frames, variadic calls, or FFI register rules. Not for the Rust FFI binding layer: use rust-ffi.
The English summary is being prepared. Showing the original description.
Use when configuring ADC sampling time, DMA-driven ADC, calibration, or DAC channel setup on bare-metal MCUs. Not for the DMA stream itself: use dma-baremetal.
The English summary is being prepared. Showing the original description.
Use when creating AF_XDP sockets, configuring UMEM and XSK rings, writing an XDP redirect program, or choosing copy versus zero-copy mode. Not for full kernel bypass: use dpdk.
The English summary is being prepared. Showing the original description.
Use when a completed session needs an agent-environment retrospective. Not for an engineering retrospective from telemetry: use engineering-retrospective.
The English summary is being prepared. Showing the original description.
Use when a redacted, trimmed agent transcript must be appended to a GitHub PR or issue body, with human approval and preview. Not for automated or model-initiated insertion.
The English summary is being prepared. Showing the original description.
Use when a repo needs agent setup, AGENTS.md added or made lean, CLAUDE.md audited, or agent instructions scored or pruned. Not for remote, credential, publish, deploy, or irreversible changes.
The English summary is being prepared. Showing the original description.