sovereign-sidecar

Agentic citation intelligence and AI-search optimization for long-form publication. Trigger on /citation-intel, "is this citation ready", "what schema for", "optimize for citation", or any request to check discoverability / schema / token efficiency / citation readiness before publication. CENTROID: article-specific citation intelligence — what makes AI agents cite this article in 2026, not generic SEO IS: - extractable-claim check (is the primary claim in the first 500 tokens?) - entity-density scan (proper nouns + concrete stats — agents cite 3× more) - FAQPage schema generator (AI cites FAQ content heavily) - community-language alignment (Perplexity cites Reddit at 24%) - token-efficiency audit (posts <10k, articles <20k, centroid <15k) - site-level AI-access stack audit (robots.txt allowlist, llms.txt) IS NOT: collapse_zones: - prestige SEO advice (keyword-stuffed listicles are tier-3, skipped entirely) - generic content marketing (must be article-specific, never templated) - llms.txt eva

Closure inference and response-geometry disclosure — direction-agnostic, scope-aware detection of materially missing expressions around an active move. CENTROID: closure inference at the point of an active move IS: - post-landing closure check (detect non-local incompleteness after a move lands) - origin-shape detection (dual-ray structure at response formulation, before commit) - surface/defer/suppress decision routing via the structural-relevance gate IS NOT: collapse_zones: - follow-up helper (complement does not generate next-steps unprompted) - documentation helper (complement does not restructure existing content) - decorative expander (must willingly say "current focus is sufficient" more often than surface) - autonomous widener (origin-shape yields shaping choice to the user; never commits the geometry) - /counter (complement maps what's MISSING; counter argues what's WRONG — different verbs entirely) sibling_overlaps: - /counter (both fire at move boun

Adversarial falsification of an active move before it commits. Fire when an irreversible decision is pending — architecture choices, schema changes, API contracts, publication, deployment. CENTROID: adversarial falsification — what's the strongest case AGAINST this move? IS: - the pre-commit adversarial stance - the falsification primitive (Popper-style: try to break the conjecture) - the sycophancy circuit-breaker (forces the model to argue against itself) IS NOT: collapse_zones: - /complement (complement maps what's MISSING; counter argues what's WRONG — different verbs) - /best-of-n (parallel implementations; counter attacks a single position) - /check (verification of correctness; counter attacks premises, not implementation) - devil's advocate theater (counter must genuinely try to break it, not perform opposition) - risk assessment (risk lists possibilities; counter constructs the strongest actual argument) sibling_overlaps: - /complement (both fire pre-mutation; complemen

Govern the fan-out to subagents — ensure the dispatch is a coherent spec, not an ad-hoc spray, before anything is spawned. CENTROID: orchestrator-as-delegator gate — govern the swarm/tranche spec before dispatch. IS: - the orchestration checkpoint: brief quality, slice soundness, synthesis contract - the Decomposition Stress Test: naming what the chosen cut GETS WRONG, not just what it correctly excludes - the verify-don't-trust-done contract: the parent's independent verification protocol for each subagent's result - posture + ingest scoping for each subagent before any work begins IS NOT: collapse_zones: - swarm executor (delegate governs the spec; spawning is downstream) - brief author (delegate governs brief quality; content is the orchestrator's) - synthesis agent (delegate specifies how verification happens; it does not do the cross-agent verification — that is the parent's execution responsibility) - scope auditor (delegate does not audit individual subagent scopes for

Metabolize external context into the working frame without cargo-culting. Fire when external content enters the session — URLs, pasted docs, API references, competitor analysis, framework docs, tutorial code. CENTROID: metabolize external context — consume primitives, retain sovereignty IS: - the gate between external knowledge and internal working frame - the anti-cargo-cult primitive (absorb patterns, reject wholesale adoption) - the sovereignty boundary (what enters must be shaped, not copied) IS NOT: collapse_zones: - /consolidate (consolidate packages known surfaces; ingest metabolizes unknown ones) - /tactical-hydration (RTCH discovers internal surfaces; ingest metabolizes external ones) - copy-paste with commentary (ingest transforms, not annotates) - summarizer (summaries lose structure; ingest preserves structure while changing frame) sibling_overlaps: - /tom (both transform; ingest transforms frame, tom transforms expression) - /tactical-hydration (both discover; RTCH

Working-mode declaration and runtime contract — declare ENG/OPS × DIRECT/META posture at session start; enforce read-only semantics for META; surface scope-overreach warnings for DIRECT. CENTROID: session-mode contract that shapes which actions are allowed before mutation IS: - one-line POSTURE banner at session start - read-only enforcement for META postures (no file writes, no commits, no API mutations) - scope-bounded mutation for DIRECT postures (only files explicitly required) - mid-session toggle surface ([Posture → X/Y]) - workspace-default fallback when posture is undeclared IS NOT: collapse_zones: - permission system (posture is contract, not enforcement; harness gates enforce — this skill describes the contract) - approval queue (posture declares intent; approvals gate actions) - audit log (the banner is observable; audit lives elsewhere) - mode coercion (the user declares; the skill describes what each declaration means) sibling_overlaps: - /counter (both pre-

Before mutation, detect whether the agent has enough local truth to act safely. CENTROID: repo-native pre-mutation gate — does the agent know enough to proceed? IS: - the practical pre-mutation readiness check (not philosophical, not post-move) - blast-radius assessment against what was actually read, not assumed - invariant-at-risk identification before a change touches a live surface - a MUTATION RECEIPT the agent produces before touching anything IS NOT: collapse_zones: - analysis theater (preflight does not produce a report about the change; it produces a receipt that GATES the change — verdict is load-bearing) - risk list (risk enumerates possibilities; preflight names the ONE invariant most likely to break and demands a validation command that would catch it) - documentation helper (preflight does not restructure or document; it clears or holds a single mutation) sibling_overlaps: - /counter (counter argues AGAINST a move philosophically; preflight asks whe

Theory of Mind (centroid-preserving) re-expression engine. Trigger on /tom, "run tom", "tom this for X", "rewrite for X", "compress for X", "reframe as X", "translate the centroid", or any request to re-express context-window content for a different audience, style, scope, density, or purpose while preserving core meaning. Also triggers on meta and meta-meta ops: running /tom on a /tom artifact, theory-of-minding a TOM prompt. CENTROID: re-express content from one expression into another while preserving its centroid (invariant meaning), source-of-truth boundaries, and audience-critical distinctions IS: - centroid extractor (what does the source actually MEAN at its core?) - posture-shaped re-expression across nine axes (audience / style / scope / density / structure / fidelity / purpose / format / diagrams) - source-of-truth boundary preserver (warnings and unresolved risks are NOT softened to fit a friendlier register) - one-turn config primitive (propose-with-alternatives, then `go` is enough) IS NO

Context consolidation — concatenate any file surface into a single LLM-consumable indexed markdown dump. Works with local dirs, glob patterns, git repos (ours or public), and conversational intent targets. CENTROID: file-surface to single-indexed-markdown-dump consolidation IS: - local directory / glob consolidation - git repo (ours or public) clone + consolidation - conversational intent resolution (grep / glob to files) - git diff range consolidation - harpoon target prep with anchor-assessment preamble IS NOT: collapse_zones: - file authoring (consolidate reads and concatenates; never writes to source) - doctrine judgment (consolidate packages for other agents; never judges content) - opinionated content filter (skip-binary and exclude-pattern are rules, not curation) - lossy compressor (truncation at limits is a transparency boundary, not a curation choice) - /tactical-hydration rewrite (RTCH selects; consolidate packages — distinct boundaries) sibling_ov

Runtime tactical context hydration — staged discovery and bounded source-bearing hydration. Answers "how does the agent know where to look before it already knows where to look?" via filesystem shape, structural signals, and typed candidate baskets. CENTROID: intent → bounded, source-reenterable evidence packet via staged source-bearing discovery IS: - structured intake of agent intent (goal, seeds, profile, fanout, mutation risk) - zone orientation (cwd / repo root / obvious truth files) - low-cost shape scout (directory map, headings, durable handles, refs) - typed candidate basket with origin / use taxonomy - tactical probe plan (multiple bounded probes, not one giant regex) - bounded chunk hydration with line-range provenance and next-re-entry commands - agent-ready evidence packet (selected_surfaces, unresolved_questions, caution_map) IS NOT: collapse_zones: - vector database (no embedding-space retrieval) - semantic oracle (does not "understand" content; surfaces structural signals) - d