Skills

Use when checking another agent's work before it ships. Runs a read-only reviewer and critic check that ends in one verdict line.

Use when creating, arming, tightening, or retiring a bot. Covers a multi-bot Hermes or Grok fleet. Sets outcome, instruction, and approval rules.

Use when Claude Code leads an Agent Teams swarm. Covers a parallel swarm on a multi-module task. Sets the spawn contract, file ownership, and proof fields.

Use when choosing or switching the Claude Code CLI model. Covers pinning, effort level, fallback, and exposing extra models through a local bridge.

Use when driving the OpenAI Codex CLI on a local repo. Covers exec flags, sandbox modes, timeouts, and checking changes that landed.

Use when running Cursor's cursor-agent CLI headless. Covers code inspection or edits, login, print-mode flags, sandbox limits, and host-side verification.

Use when reviewing another agent's data deliverable. Covers CSV, JSON and other data outputs, such as a list, extract, or catalog audit, before a merge. Verifies claims by re-execution.

Use when a delegated worker's output must be checked. An orchestrator confirms it is real work, not a stub. Reads files on disk and returns a REAL, STUB, or SUSPECT verdict.

Use when claiming a background job is running or done. Requires a process ID, a fresh process check, and read output before any status report.

Use when delegating repo edits to a headless coding CLI. The orchestrator keeps scope, safety, review, and verification.

Use when setting up a free-tier mixture-of-agents preset. Draws on free-tier model providers through a local OpenAI-compatible router. Covers install, live model discovery, and preset setup.

Use when driving the Grok Build CLI on a named branch. Runs with its memory feature on. Covers memory setup, prompt ordering, interrupted runs, and safe commits.

Use when writing a work order or task card for an agent. Applies to coding harnesses too. Sets the card shape, pinned model, done criteria, and review handoff.

Use when dispatching a kanban board to coding harnesses. An orchestrator runs this for harnesses such as Claude Code, Codex, and Grok Build. It covers lane routing, worktree isolation, the independent judge gate, and the card completion contract.

Use when designing or maintaining role routing. Covers runnable agent profiles, identity documents, and provider fallback chains for a multi-agent Hermes fleet.

Use when a kanban task spawns coding or analysis work. Covers architecture work too. Writes a spec file first, then runs readiness and course-correct gates.

Use when briefing a read-only reviewer on a code path. Covers a second, read-only session that audits a composed path, such as checkout, payment, or access gates, for defects that no single diff shows.

Use when managing tasks on a kanban board. Covers creating, dispatching, linking, checking status, and diagnosing a Hermes kanban run that keeps cycling.

Use when creating, blocking, or unblocking a kanban card. Covers splitting a card, and tasks that look stalled, duplicated, or orphaned on the board.

Use when several agent workers share one kanban goal. Covers card contracts, dependency links, dispatch checks, review gates, blocker routing, and recovery after crashes or timeouts.

Use when a kanban worker crashes or gives up early. Covers failures within about two minutes of starting. Matches the failure to a cause before any repair runs.

Use when installing or verifying a local agent CLI. Covers authenticating and updating it, for a CLI that works inside a project directory. Ends with a live model response, not an install command.

Use when two agents' branches conflict in a git merge. It also covers a rebase, where neither agent should decide the outcome alone. A neutral third party reconciles the conflict.

Use when parking a titled agent session for later resume. Creates or reuses a titled session as a parked slot. Starts the slot without doing any work.

Use when handing a long run to an Opus model. Applies to a long coding run, a design or copy task, or a pre-review pass in Claude Code or Claude apps. Covers the task brief, effort, stop rules, and checking results.

Use when several agents share one group chat. Each message must name the agent it is for. Covers the addressing prefix, reply routing, and when to stay silent or ask a question.

Use when an adversarial review returns REQUEST_CHANGES. Findings must be verified before any rework. Confirms or dismisses each finding first.

Use when a dispatched work order ends in a gated merge. Covers reading the order first, pinning the head SHA, merging only the reviewed commit, verifying the merge in the tree, and keeping the results ledger intact.

Use when a capability is missing or a claim is unproven. It also applies when a result looks empty. Report the constraint plainly and immediately, with no optimistic wording.

Use when an instruction could mean several things. It applies when the candidates are several sources, targets or scopes. Find every candidate first, then ask before planning, unless only one candidate exists.

Use when driving a desktop GUI with several windows open. Confirm the target window is in front and the input reached its field, and hand the keyboard back when it did not.

Use when answering config or vendor questions. The same applies to keys, UI, how-to steps or architecture. Load an authoritative source before answering or acting, because a guessed answer on these topics is a hard failure.

Use when an agent loops or narrates without acting. It also applies when a running loop must stop. Set read limits, pair every progress claim with a change, and halt processes so they stay halted.

Use when starting any coding or repository task. Query the code-graph server for the touched area before reading repository files, check freshness, and if the graph is unreachable or stale, say so and continue.

Use when planning or reviewing a code or config change. It also covers proposing such a change. Choose the smallest verified, reversible change, label every claim by evidence, and stop at each external side effect for explicit approval.

Use when collecting facts without changing anything. It covers files, git history and running processes. Search in a fixed order, record metadata for every hit, mark what could not be proven as unknown, and keep secrets out of the report.

Use when telling the operator something is done or live. It applies to anything that will happen, is done, is live, or was created. Run the check in the same turn, quote its output, and correct any earlier claim the check does not support.

Use when asked how to remove an installed tool. It also covers what the tool does and where it writes. Read the upstream installer source first, list the install footprint and its reseed sources, then ship a dry-run removal that backs up before it deletes.

Use when naming a vendor, brand or sender in a reply. It also covers a corporate parent, a product family, a task record, a pull request or a message. Verify the corporate relationship with one cheap probe first, and do not call a sender suspicious from a pattern match alone.

Use when wiring a Claude CLI bridge into Hermes. The bridge is wired in as a provider. Covers bridge install, custom provider registration, launchd keepalive, and OAuth re-login.

Use when authoring or debugging a cron job. The job runs CI, repo watches or other automation. Covers toolsets, silent failures, honest reporting and escalation.

Use when a profile's config.yaml will not parse. It also applies when a model change lands in the wrong profile. Covers scope checks, YAML repair, schema checks and restart verification.

Use when dispatching or steering an agent run. The run is dispatched over the Hermes local API. Covers POST, polling, steering, stopping and wedge triage.

Use when debugging or changing Hermes Agent internals. Covers the module map, config and profile layout, subsystem pitfalls and how to verify claims upstream.

Use when a Hermes provider login or key fails. The login or key may look expired. Covers the runtime credential check, credential pool cleanup and env-versus-pool confusion.

Use when auditing a Hermes profile config snapshot. The snapshot is read-only. Covers the fields to capture, what empty fallbacks mean and how to cross-check auth.

Use when reverting a Hermes profile's model to a backup. Later changes are kept. Covers snapshot, block-only restore, verification and reporting.

Use when changing Hermes toolsets or MCP servers. The change is made in a profile config. Covers live-config checks, CLI limits, list-valued keys and operator-approved edits.

Use when rolling out or auditing Hermes fallback chains. The chains span several profiles. Covers the chain rules, credential gaps, verified writes, watching for silent fallbacks and rollback.

Use when a Hermes gateway's first model call is slow. It takes about 30 seconds after a restart. Covers the cold-path cause, how to measure it and when a fix is worth it.

Use when a Hermes gateway stops responding. It also applies when Telegram polling fails. Covers service identity, logs, restart discipline, failure signatures and health checks.

Use when a Hermes issue needs an upstream check. Check it before you report, file or change anything. Covers upstream verdicts, drift and safe apply.

Use when the Hermes model picker shows wrong models. It shows zero, duplicate or stale models. Covers config-versus-session truth, picker data formats, cache refresh and picker implementation bugs.

Use when wiring a model provider or fallback chain. It also applies to auxiliary lanes and to sweeping a retired model from every route. Covers slugs, lanes, pools and quota signals.

Use when creating a Hermes profile. It also applies when copying its skills and MCP servers. Covers fresh creation, selective skill copies, per-profile wiring and kanban pins.

Use when publishing a Hermes skill to GitHub or a hub. Publishing goes to a public GitHub repo and a community hub. Covers account checks, credit, install URL proof and the hub submission.

Use when a user needs the right slash command. The task is already running. Covers the decision rule for steer, queue, background and interrupt, plus goal, loop and plan.

Use when filing an issue or pull request upstream. The target is the Hermes upstream project. Covers ruling out local causes, deduplication, issue-first scope, small diffs and evidence.

Use when a Hermes web UI must be reachable from a phone. The route runs over a private network. Covers a private-address bind, the daemon start, a debug ladder and separate SSH checks.

Use when a CLI wizard waits for menu input. You must provide that input yourself. Covers PTY launch, arrow-key menus, polling between steps and checking the result.

Use when installing or repairing a Hermes context plugin. It is a lossless context-engine plugin. Covers the three activation steps, registration checks and compaction tuning.

Use when setting up Hermes with several Grok roles. The roles use xAI OAuth. Covers role-to-model mapping, delegation limits, memory settings and profiles.

Use when a Hermes user wants an animated mascot. It can run in the terminal or on the desktop. Covers browsing, installing, selecting, sizing and diagnosing pets.

Use when terminal commands fail on a removed directory. It also applies before any change to a repo you may not be in. Covers stale working directories, retry loops and preflight checks.

Use when choosing routes for expiring subscription quota. Covers evidence labels, per-window telemetry, Hermes-native routing and canary changes.

Use when a script is about to write a Hermes config file. Covers the verification ladder that proves a write is needed and correct.

Use when Hermes must use xAI Grok by subscription login. It is used instead of an API key. Covers the browser login, model selection, fallback order and login troubleshooting.

Use when installing or fixing the Clude memory sidecar. It covers its MCP server and dream or reflection daemon, beside an existing memory system. Clude is additive and must not replace the memory systems already in place.

Use when deploying the Eidolon layer on Hermes. Its scope covers the SOUL.md identity file, the dream-cycle handler, the integrity watchdog, scheduled reports, and the operator-policy rules that govern its changes.

Use when diagnosing Graphiti timeouts or pool pressure. It also applies to hangs, keepalive drops and writes that fail with missing credentials. Covers the tuning ladder and the event-loop and embedder traps.

Use when installing or debugging Hermes memory providers. They include Hindsight, Omega, Graphiti, or a bridge across them. Covers the setup wizard, key slots, per-profile isolation, retain and recall proof, and a triage ladder for failed writes.

Use when installing cloud-mode Hindsight coding agents. The goal is that every supported harness writes to one designated memory bank. Captures a baseline, pins routing keys, backs up harness configs, and verifies that no extra bank or mental model appeared.

Use when storing or retrieving long-term agent memories. It also covers the Hindsight daemon, retain failures, cloud or local mode, and bridges to another store.

Use when snapshotting a Hermes memory stack. It covers profile state, Hindsight native state, the Omega SQLite store, the Graphiti MCP configuration and source, and the Neo4j graph, and restoring is covered too. Verify each layer before relying on the snapshot.

Use when distilling large official docs. The references keep exact endpoints, headers, signatures, imports, paths and commands. Search first, read bounded regions, and verify every file and the final response.

Use when Hermes runs three bridged memory stores. They are a primary fact store, a local semantic index and a temporal graph. Route each read and write to the right store, keep the stores consistent with an idempotent bridge, and report partial writes honestly.

Use when routing Claude Code to local models on a Mac. Covers a stripping proxy, three-tier model routing, per-model server settings, and preflight checks.

Use when running or re-modeling Darkbloom provider Macs. Covers update recovery over SSH, stale PID fixes, revenue-driven model selection, safe model switches, and a controller-agent pattern.

Use when onboarding a Mac as a Darkbloom provider node. Covers install, device link, MDM, model pinning, daemon start, and the done gate.

Use when two coding agents share a local DeepSeek engine. The engine runs on one Mac. Covers worktree isolation, role split, approval gates, and config for each client.

Use when operating the DS4 inference server on macOS. It serves DeepSeek models. Covers launchd setup, health checks, safe restarts, model switches, and multi-Mac engine updates.

Use when building a distributed MLX inference cluster. It runs across two or more Apple Silicon Macs. Covers install, direct-link networking, storage, model placement, and recovery.

Use when picking MLX models per EXO memory tier. Covers discovery, size bands, same-band benchmarking, champion promotion, and report evidence rules for an EXO cluster.

Use when registering uncensored MLX models on EXO. It applies to an EXO cluster. Covers curated discovery, RAM-budget scoring, API registration, download, verification, and state tracking.

Use when fetching large model weights from Hugging Face. Covers token auth, resumable single-file downloads, per-file snapshots, and persistent background runs.

Use when running, debugging, or retiring LLM servers. They run locally on a Mac. Covers port coexistence, single-model residency, auth and bind rules, memory-guard errors, client routing, and safe model decommissioning.

Use when running or tuning local MLX inference servers. They run on a Mac. Covers safe config edits, restarts, per-model settings, side-by-side variants, and single-GPU sharing.

Use when preparing a Hugging Face model for MLX. It is for MLX inference. Covers source verification, resumable download, shard repair, quantization choices, and benchmark gating.

Use when adding, converting, or recovering oMLX models. Registration runs on an oMLX server. Covers download, conversion, admin-API registration, sleep and wake recovery, and cache placement.

Use when tuning one oMLX model's settings. It also covers validating, promoting, and rolling back a change on an oMLX server. Covers schema-first edits, read-back checks, promotion gates, and benchmarking.

Use when a Mac's disk or memory is under pressure. Covers a read-only audit, safe cleanup order, halt conditions for local model servers, and reporting reclaimed space.

Use when operating a Hermes agent host day to day. Covers safe updates, backed-up config edits, provider fallback chains, credential handling, and checking which file a live service reads.

Use when adding a second Mac as a Hermes agent host. Covers private-network SSH, the agent install, profile and bot setup, the gateway service, delegation, and verification.

Use when a process quit, died, or restarted on its own. The cause must be found from evidence. Covers timelines, ranked elimination, probe hygiene, and an honest report.

Use when a scheduled agent watches Mac services. It watches critical services on two or more Macs, reports config drift, and flags upstream releases without changing operator-owned config.

Use when a cutover left stale host and endpoint names. Stale references can include addresses. Covers live-versus-evidence triage, backups, a re-scan gate, and reporting.

Use when restoring Screen Sharing or VNC access to a Mac. Covers mode conflicts, socket-activated listeners, authentication isolation, private-network access, and verification.

Use when reading live Apple Silicon metrics. Covers CPU, GPU, memory and thermal data, the interactive view, one-shot JSON for telemetry, and export.

Use when working across several Macs in one fleet. Covers host identity, live reachability probes, safe cross-host commands, service restarts, and fleet-wide config changes.

Use when running a scheduled overnight health pass. It is a health and hygiene pass on a Hermes agent host. Covers gateway health, memory store hygiene, skills and cron checks, config review, tool dependencies, and backups.

Use when Spotlight or model servers strain memory. The memory or disk pressure can come from Spotlight indexing or from two model servers sharing one memory pool. Covers indexing toggles, single-server rules, and diagnosis.

Use when attaching an agent to Tailscale Aperture. Aperture is a gateway. Covers MCP connector setup, least-privilege grants, and tailnet-only inference routing.

Use when a Tailscale node on a Mac fleet needs recovery. Covers key expiry, Tailnet Lock signing, ACL reachability, node removal, variant choice, and MagicDNS rules.

Use when a Hermes host reads or writes Bitwarden secrets. It uses the Bitwarden Secrets Manager CLI for credentials. Covers machine accounts, scoped read grants, no-echo handling, upsert and verification, and disk-key consolidation.

Use when a secret must reach an env file, not chat. Covers operator-entered staging, file permissions, gitignore-first, fingerprint-only verification, and clipboard hygiene.

Use when pushing from a repo or checking a remote URL. A remote URL may carry a token. Covers redacted remote checks, credential helpers, explicit push refspecs, and token rotation.

Use when a GitHub repo or skill needs a preflight scan. An MCP server must be checked the same way before it is installed or run. Covers quarantine, classification, scanner order, conservative findings, and hard stop conditions.

Use when a key must reach only some Hermes profiles. It must reach them through the secrets command helper, not bulk Bitwarden injection. Covers the helper script, per-profile config, override safety, and fingerprint checks against the store.

Use when a local inference server's key must be rotated. Rotation follows exposure or a maintenance schedule. Covers finding every consumer by hash, atomic updates, restarting only the server, and proving the new key works while the old one is rejected.

Use when injected text surfaces in cron or chat output. Covers treating injected text as data, redacting config before output, tracing the ingest path, containing the job, and hardening the rebuild.

Use when auditing plaintext secrets on a host. It also migrates them to a shared Bitwarden Secrets Manager project. Covers inventory without reading values, cross-referencing the store, updating consumers, read-only surface checks, and shredding only after operator approval.

Use when a change touches a compliance gate. The question is whether a fail-closed gate still blocks, such as a restriction at checkout, an age check, or a blocklist. Covers a read-only review, transpiled probes, data reconciliation, surface scans, and a verdict file.

Use when deploying a static site to Cloudflare Pages. The deploy runs headlessly. Covers a token-safe deploy script, a content-verification gate that an HTTP 200 alone does not satisfy, custom-domain order, cache busting, and rollback from a backup.

Use when building a Cloudflare Worker with KV storage. It has a cron snapshot and sits behind Cloudflare Access for public-internet users. Covers server-side secrets, disabling preview hosts, JWT verification, free-plan limits, fail-visible data, and verification.

Use when reviewed docs PRs should land as one draft. Several open, reviewed, docs-only pull requests are combined into one draft pull request. Covers snapshotting live state, skip rules, per-PR commits copied from each source head, gates before push, and a verification block for the description.

Use when running or porting an exported design bundle. A bundle is a zip of front-end source with a preview host. It may run locally or move its scene into a storefront. Covers serving over HTTP, reconciling asset paths, porting to Next.js, and layering 3D content under page text.

Use when carrying a GitHub issue to a verified PR. It reports CI honestly. Covers reading the full thread, sweeping for duplicate work, validating the premise against design intent, fixing the bug class, proving the regression test, and reporting CI from live evidence.

Use when Medusa Cloud auth expires or a build is needed. It also triggers storefront or backend builds and proves a deploy runs a specific commit. Covers auth refresh, build triggers, commit verification, tracking branches, variables, domains, backups, and triage.

Use when installing or auditing the OpenWarp fork. It is an open-source terminal fork that installs and connects to a local model server on a Mac. Covers identifying the right binary in the bundle, configuring a local provider, auditing outbound connections, building from a pinned commit, and MCP configuration.

Use when verifying a pull request's head files. Checks hashes, sizes, existence and branch state without a clone, or runs a repository's acceptance script against a PR head read-only.

Use when telling anyone that a change is live or fixed. Proves that the production build contains the change, using the deployment record, the bundles the production page loads, and live behavior.

Use when backing up a git repository before risky work. Makes tarballs with checksums, runs a secret scan, pushes to a private backup remote and verifies each ref.

Use when repo docs or remote git state mislead agents. Keeps one current-truth document accurate and reconciles remote state with local history.

Use when publishing code or docs publicly on GitHub. It must carry no personal identifiers, machine identity, network details or secrets. Covers maintaining that public repository afterwards.

Use when preparing a commit or PR for production code. Pause after the first working draft, verify the external symbols, trim the scope, then ship.

Use when auditing a repo for AI code quality. Covers auditing or fixing a GitHub repository against the Slophammer standard for testing gaps, dependency boundaries, complexity, type safety and coverage gates.

Use when publishing a Medusa storefront or backend. Pins the approved commit, proves the live build, and runs layered checks before calling the release live.

Use when making a website readable to search and AI. Covers crawlability, structured data, agent files, accessibility for agent browsers, bot access and a verification gate, without changing what human visitors see.

Use when auditing whether AI answers cite a brand. Records who is cited, labels every surface that could not be observed as UNKNOWN, and delivers a matrix, raw observations and a report as a draft pull request.

Use when building or auditing a marketing page. Scores the page against 30 common tells, and passes it only when no more than five are present.

Use when a live app needs a scripted smoke test. The interactive browser cannot authenticate or capture network evidence. Covers headless runs, request capture, redaction and blockers.

Use when auditing a storefront's conversion and UX. Read-only, ranked backlog with evidence per row.

Use when auditing a live storefront for broken links. Read-only crawl, per-row CSV, draft PR.

Use when measuring a live storefront's page speed. Uses Lighthouse and ships the baseline as a docs-only draft PR.

Use when running a QA matrix on a preview storefront. Playwright runner, JSON evidence, generated report.

Use when auditing storefront WCAG 2.2 AA accessibility. Read-only, per-rule CSV, draft PR.

Use when auditing a storefront's source for a pattern. Read-only, per-row CSV of file:line findings, report, draft PR.

Use when accepting storefront visual changes. Covers fixed-canvas heroes over CSS backdrops, scroll-driven entrances, product grids and commit-pinned preview evidence.

Use when a storefront page adds or changes a control. Match the sibling controls on the same page.

Use when auditing a storefront catalog for drift. Produces a per-row CSV of defects and a named-rule report, without changing live data.

Use when a catalog product image is missing or wrong. Covers search, live URL checks, vision verification and an honest residual list.

Use when turning product photos into catalog media. Covers a read-only preflight audit, barcode grouping, OCR, gallery replacement and read-back checks.

Use when preflighting catalog writes or price changes. Proves identity, sibling coverage, sell-unit pricing and compliance gates before any write, then requires a read-back check.

Use when batch-processing storefront product photos. Covers 3D assets, pruning a shoot folder to hero shots, watching a shared photo folder, and a visual operator gate before anything reaches the catalog.

Use when wiring opt-ins and opt-outs to email services. Covers a durable suppression ledger, event-time ordering and a five-case test matrix, so that an opt-out always wins.

Use when building a storefront readiness checker. Covers the lesson record format, portable checks, consolidation timing and checking a new site against known incidents.

Use when checking live feeds and structured data. Read-only, re-runnable, with a self-test.

Use when a Medusa admin page crashes on orphaned rows. Covers diagnosis, a fail-closed SQL repair with backups and tested sync patterns.

Use when orchestrating Medusa.js work with coding agents. Covers harness roles, official docs and agent plugins, a B2C-first monorepo boot, vendored compliance-module wiring and bank-gated payment sequencing.

Use when bridging Zoho Inventory and a Medusa store. Includes read-only stock digs against a Zoho dump.

Use when verifying a Medusa backend over HTTP. Covers token classes, read-only probes, publishable-key discovery, product counts and deploy-SHA gates, without admin keys or test passwords.

Use when writing or changing Medusa code and API calls. Requires the installed version, a cited official page or the installed types for each version-sensitive shape, and a stop with one exact question when the docs are silent.

Use when changing Medusa catalog or inventory data. Covers auth, relation expansion, read-back after every write, safe large reads and the variant-image link rules.

Use when writing product descriptions from sources. Every claim is cited, a two-pass voice audit runs, and facts that cannot be sourced become open questions.

Use when adding state, age or license gates to a store. Covers one legal blocklist, fail-closed checks, cache-stable pages, a dark-launch arming sequence and verification by hash.

Use when drafting copy for a regulated online store. Output is one draft pull request, never merged, with a compliance lint that must report zero hits.

Use when verifying certificates of analysis (COAs). Covers source discovery, identity checks against the document text, status codes, wall hosts and a metadata bind with read-back proof.

Use when building a research-use-only storefront. Covers copy rules, visual proof and the dark-not-delete rule for compliance changes.

Use when running a read-only wording compliance sweep. Crawls every public page with GET requests, flags copy that breaks research-use-only or similar rules, and produces per-hit evidence, a rewrite for each flag and a report generated from the data.

Use when building a sourced question bank for AI search. Covers buyer questions for answer-engine coverage, mapped to a site's live pages.

Use when generating AI brand video or image content. Covers regulated or character-led brands whose logo and characters must stay fixed.

Use when comparing competitor storefronts by feature. Every cell cites a fetched URL or says UNKNOWN; output is a draft PR.

Use when checking claims against official sources. Uses verbatim quotes a script can re-fetch and a per-claim verdict.

Use when running deep research in Perplexity desktop. It runs through a local vision-based desktop agent (HoloDesktop CLI) instead of the web.

Use when building a verified list of business prospects. It is deduplicated, drawn from public registries and map data, and delivered as a CSV with a source link on every row.

Use when drafting storefront copy as files for review. Every claim cites a live page or source file.

Use when publishing any outbound storefront link. Checks UTM parameters against the team's naming convention and blocks links that fail.

Use when operating the Bento email platform. Work through its API, MCP server or dashboard. Covers keys and site scoping, which surface to use, safe-send rules and common pitfalls.

Use when a task needs browser control on macOS. Covers choosing between agent, managed-session and scripted approaches, reliable form and frame handling, and guardrails.

Use when setting up a self-hosted Buzz relay. It also covers debugging with Docker Compose. Covers the hostname rules, membership roster, domain changes, layered agent debugging and backups.

Use when a Hermes gateway reaches a remote Buzz relay. The gateway runs on one machine and the relay is used from another. Covers the identity rules, admission checks, the outbound CLI, verification and media attachments.

Use when an agent sends or reads Slack messages directly. The route is the Slack Web API. Covers token handling, mention IDs, thread replies and the limits of bot-to-bot wakeups.

Use when connecting an MCP server to a Hermes profile. The profile uses the server as a client. Covers pre-flight probes, the config block, the session boundary and smoke tests.

Use when building a small Python MCP server from a spec. Its tools are wired to a fetch or scrape function. Covers the runtime, library version traps, FastMCP wiring and config.

Use when agents in one Telegram group start a loop. It also applies to floods and repeated messages. Covers the causes, log checks, consolidated replies and fixes.

Use when connecting an agent to n8n Cloud. The connection uses the instance-level MCP server. Covers OAuth, tool whitelists, the fixture-proof gate and human approval for publishing or executing.

Use when you need live model benchmarks or pricing. The data also covers provider latency and credit balance, from the OpenRouter MCP server. Covers tools, billing, routing suffixes, tokens and a REST fallback.

Use when probing a stdio MCP server with no client. The server has no client set up yet. Covers the handshake, the tool list, evidence records and the silent-server case.

Use when scraping through a self-hosted Firecrawl server. The server runs on a private network, and search and crawl jobs are also covered. Covers the health check, page scrape, search, crawl jobs and safe-use rules.

Use when a Hermes Telegram bot needs large-file handling. It covers reactions and the optional local Bot API, including troubleshooting. Covers the limits, the config keys and the agent behavior rules.

Use when connecting TinyFish search to coding agents. It applies to search and fetch. Covers key handling, unattended connection, the doctor check and a search-and-fetch proof.