agentic-core

Plugin agentic-core from marketplace agentic-core in baristaze/agentic_core. Commands read from the repo's manifest — 19 days ago.

Skills (41)

• audit-credential-lifetimes

  Audit how long each credential keeps working after it should stop: one row per credential kind (auth session, API key, operator token, sign-in, socket ticket, and any other the code has) and channel (HTTP request, realtime socket, bus control message, worker), with where it is checked, how often, wh

• audit-database-calls

  Audit how many database calls each endpoint and each worker flow makes, at its least and at its most: round trips, transactions, and roles, counted at the driver while the real API and worker run in one process over a database of their own. Names what grows a flow, every N+1, every read repeated wit

• audit-deploy-time

  Audit where a deploy's minutes go: the pipeline's jobs and steps from the Actions run, the Terraform apply's resources, the migration's one-off task, each service's rollout from its ECS events, its tasks, and their first log lines, and the settings that pace them (health-check grace, container healt

• audit-model-spend

  Audit what an org's model calls spend, in one environment, from the usage records the engine wrote, read through the operator plane's usage route with a read-only operator token for the sessions named: the spend by agent kind, kind version, model role, and model, the prompt tokens read from a cache,

• audit-provider-calls

  Audit every call to an external provider (WorkOS, the AWS services, any HTTP to a third party), per flow: each route, inbound webhook, worker job, and boot. For each flow, the calls in order, whether they repeat or stand apart, whether they sit on the request path, how often the flow runs, whether t

• audit-query-indexes

  Audit whether the indexes fit the queries, and what breaks first under load: every statement the Postgres storage sends, mapped to the index that serves it and measured with EXPLAIN on a database of its own seeded at a stated scale, under the login and the tenant scope the application uses. Reports

• audit-retention

  Audit which stores grow without bound and what trims them: every table of every role, with what writes it, when a row is stale, the purge that deletes it, its retention, and whether that purge is batched and indexed; then the stores outside the database (log groups, buckets, queues, the cache). Meas

• docs-compact

  Compact the tree's documents to what holds at the head of the main branch: rewrite each ADR to its present decision with a status of one date, remove the ADRs that constrain nothing, cut the changelog to the latest release, clean the rows of specs/architecture.md, and rewrite the comments that tell

• ops-agent-session-spend

  Reconcile one session's model spend, in one environment, loop by loop, from the usage records the engine wrote: the session's rollup and each loop's, read once through the operator plane's usage route with a read-only operator token, then ops-loop-spend's reading over each loop, at most ten. Takes t

• ops-cloud-deployment-create

  Create one cloud environment of the system from nothing, in its own AWS account, as that account's administrator: the bootstrap root (state bucket, registry, OIDC trust, the deploy roles, the investigate role, the budget, the zones), the delegation of its public names at Cloudflare, the investigate

• ops-cloud-deployment-nuke

  Destroy one cloud environment of the system as its account's administrator: empty the buckets, destroy the environment root, remove what Terraform does not own (the secrets the application wrote, whose tenants' share follows the database's final snapshot, and the leftovers AWS made), and report what

• ops-infra-as-code

  Write or change the system's Terraform in the shape the guideline prescribes: one module graph shared by every environment, a variable for everything that differs, the autoscaling switch, the destroyable switch, the budget, the alarms, and the tags; then plan it under the read-only investigate profi

• ops-integration-silent

  Say why an integration went silent, in one environment, with a read-only credential: count what its inbound webhook route answered, read what the worker made of each delivery and the lines it logged, and read the webhooks queue and its dead letters, then report where the deliveries stop, why, and wh

• ops-investigate

  Investigate one environment of the system with a read-only credential: the alarms, the error rate, the latency, the outcomes, gauges, queues, and log lines each namespace's signals file names, the pool, the cost against the budget, and the system's size, then report what is wrong and what to do next

• ops-loop-spend

  Reconcile one loop's model spend, in one environment, from the usage records the engine wrote, read through the operator plane's usage route with a read-only operator token: each turn's tokens by class and its cost, the context's growth across turns, cache-miss streaks, compaction churn, and the cos

• ops-root-cause

  Find the root cause of one tenant's problem in one environment: read that tenant's rows through the operator plane's read routes with a read-only operator token, correlate them with the logs, the trace, and the error event by request id, at most five ids and one pass each, and report the cause and t

• ops-simulate-traffic

  Drive realistic traffic at one environment of the system through its edge with the system's own traffic generator, at one of four profiles (light, regular, heavy, stress) for a duration, and report the table: requests by route and status, p50, p95, p99, and the error ratio. Use to warm an environmen

• ops-watch

  Watch one environment of the system live from a sub-agent: a log tail, the alarms as they fire, and the error and latency signals, batched per interval and capped, with a read-only credential. Run it in a sub-agent the invoking session spawns, because it polls for the whole window, at most 30 batche

• stress-test-create-or-update

  Write or edit one stress test scenario of the system, ops/stress/<name>.yaml: the profile, the duration, the ramp, the soak, the target p95 and error ratio, and the routes it weights. States what a good stress test has and refuses to write one without a target. Writes a file and nothing else; a run

• stress-test-run

  Run one stress test scenario of the system against one environment with the system's own generator, read the signals back through their own APIs, and report pass or fail against a target stated before the run: the scenario's p95 and error ratio, or the one the run states instead. A real run is the s

• tickets-triage

  Triage the open tickets whose subject is this repository: for each, read it and check it against the code on main, the merged and open pull requests, and the history, then give it a verdict (done, moot, partly done, still relevant, in flight, out of scope, unsure) with the evidence. Writes a report;

• agentic-benchmark-browser

  Run the scenario browser-judge-agentic: send its prompt, which names this repository's public URL, in the chatgpt.com, claude.ai, gemini.google.com, and grok.com products, signed in, and save each answer with its conversation URL as proof; in a checkout of this repository, also check the run in, red

• agentic-deviate

  Record a deliberate deviation from a rule of the agentic_core spec as an ADR under docs/adr/: the rule quoted, the decision, the consequences. Use when a finding of an engine review is accepted as intentional.

• agentic-explain

  Explain how the agentic_core spec applies to a question, a file, or a change, citing the sections and lenses that govern it. Use when onboarding to engine code or before a change that crosses its parts.

• agentic-review-agents

  Engine review through the Agents lenses of the agentic_core spec: Agent Kinds and Sub-Agents. For a change in this area, or as one leg of agentic-review-full.

• agentic-review-bounds

  Engine review through the Bounds lenses of the agentic_core spec: Bounds and Budgets; Parking. For a change in this area, or as one leg of agentic-review-full.

• agentic-review-economy

  Engine review through the Economy lenses of the agentic_core spec: Economy. For a change in this area, or as one leg of agentic-review-full.

• agentic-review-full

  Full engine review: the ten lens groups of the agentic_core spec run in parallel and merge into one report. Use before a pull request to engine code, or when a change crosses groups.

• agentic-review-live

  Engine review through the Live lenses of the agentic_core spec: Streams; Steering. For a change in this area, or as one leg of agentic-review-full.

• agentic-review-models

  Engine review through the Models lenses of the agentic_core spec: Models. For a change in this area, or as one leg of agentic-review-full.

• agentic-review-privacy

  Engine review through the Privacy lenses of the agentic_core spec: Privacy; Null Objects. For a change in this area, or as one leg of agentic-review-full.

• agentic-review-steps

  Engine review through the Steps lenses of the agentic_core spec: The Engine and the Brain; Steps; Loops, Runs, and Sessions; History. For a change in this area, or as one leg of agentic-review-full.

• agentic-review-tools

  Engine review through the Tools lenses of the agentic_core spec: Tools; The Runtime; Workspace Snapshots. For a change in this area, or as one leg of agentic-review-full.

• agentic-review-trust

  Engine review through the Trust lenses of the agentic_core spec: Identity, Trust, and Attribution. For a change in this area, or as one leg of agentic-review-full.

• agentic-review-windows

  Engine review through the Windows lenses of the agentic_core spec: Context. For a change in this area, or as one leg of agentic-review-full.

• agentic-scaffold-agent-kind

  Add an agent kind, or a new version of one: its tools, its done rule and result tool, its authority mode, and its tree's bounds, declared in the product's catalog and registered with every process that builds the managers, with its tests, in the shape of the engine's agent kinds. Python.

• agentic-scaffold-model-role

  Add a model role: the task a call names, its fill and its declared fallbacks in the resolver's table, a price row for every model they name, and the tests, in the shape of the engine's main and summarizer roles. Python.

• agentic-scaffold-new

  Start a new system on the engine: pick its name, copy the engine's scaffold under it, run its gates, record the product's first decisions in the thousand above its base's, then add its first agent kind.

• agentic-scaffold-provider-adapter

  Add a model provider's adapter under integrations/: the one content shape translated both ways, what does not survive named, its errors read into kinds, no retries of its own, the scripted twin standing in for it, its settings, and tests over recorded payloads, in the shape of the engine's two adapt

• agentic-scaffold-tool

  Add a tool an agent calls: its input and output, class, effect, timeout, and mode, its target and preflight, its run through the call's runtime, and its tests, in the shape of the engine's tool contract; then name it in the agent kinds that call it. Python.

• agentic-upgrade-scaffold

  Move a platform's base to a later engine release: a product rendered from the engine's scaffold, or a layer that takes its scaffold/ folder unchanged. Runs the guideline's arch-upgrade-scaffold with the engine as its source, so one move takes both foundations, and adds what the engine's layer holds:

Subagents (1)

• agentic-reviewer

  Reviews a scope of code through exactly one lens group of the agentic_core spec and returns the standard review report. Used by agentic-review-full to run the ten groups in parallel; can be delegated to directly with a group name, a scope, and the absolute paths of the lens file and the spec.

Similar plugins