agentic-os-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Capabilities
Features and capabilities supported by this server
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
| prompts | {
"listChanged": true
} |
| resources | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| search_methodologyA | Full-text search over the agentic-os governance, agentic-sdlc pipeline, and agentic-qe blueprint documentation. Use it first whenever you do not already hold a document URI, then pass a result uri to get_document for the full text. Matching is AND, not OR — every term must appear — and terms match at word starts, so "gate" finds "gates" and "gating" but "ate" finds neither. It searches markdown only: role presets are JSON and never appear here, so use list_presets for those. Read-only and idempotent; searches the server's own bundled corpus, never the network or your repository. |
| get_documentA | Fetch the full text of one agentic-os methodology document, named by its exact agentic-os:// URI. Use it to read a document you have already located — normally via search_methodology, whose every result carries the URI to pass here. It resolves exact URIs only and cannot search, so a guessed URI returns an error rather than a near match. Read-only and idempotent: nothing is ever written, and the same URI returns the same document. A body longer than max_chars comes back cut at a code-point boundary with truncated set and total_chars giving the full length. |
| list_presetsA | List the agentic-os role presets — the named role bundles (developer, qa, architect, devops, and so on) that decide which governance files an install scaffolds and how much human approval it demands. Use it to discover the valid role names before calling plan_install, or to compare what roles differ on; each entry carries a uri you can read for the preset in full. Takes no arguments and always returns every preset. Read-only and idempotent. |
| list_qe_blueprints | Browse the agentic-qe Quality Engineering blueprint catalog, organized by STLC stage. Each blueprint describes how to build one piece of QE capability (test-coverage analysis, flaky-test debugging, and so on); use this to find the blueprint that fits a QE task, then read its uri for the instructions. Returns titles and one-line summaries only — it is a catalog index, not the blueprint content. The stage argument enumerates the valid stages. Read-only and idempotent. |
| list_sdlc_phasesA | List the agentic-sdlc pipeline phases in execution order, with each phase's skip condition and the judgment gates it raises — the points where the flow must stop for a human decision. Use it to drive the SDLC flow yourself in a host that cannot run the plugin, or to answer what happens when in the pipeline. Returns the phase and gate skeleton only; read source_uri for what each phase actually does. Takes no arguments. Read-only and idempotent. |
| plan_installA | Compose one or more agentic-os role presets into an ordered manifest of the files an install should scaffold, each with a uri to read its template from. Use it after list_presets to turn chosen roles into concrete steps. Composition is additive: roles are unioned, the strictest HITL level wins, and every orchestration style in the union is installed. This returns a plan and writes nothing — no file is created, and the target repository is neither read nor touched, so the plan is not validated against what may already be installed (use run_doctor for that). You perform every write yourself, so the user can review each one. Read-only and idempotent. |
| run_doctorA | Audits an agentic-os install in a target repo you name. Runs the checks that are pure file inspection natively and returns a verdict for each; the three checks that need executing Python (hook compile+import, canned-event dry-runs, HITL smoke) come back as exact commands in host_must_run for you to run yourself — this server never executes code from a target repository, and two of those three commands sets (dry_runs, hitl_smoke) write and then delete a probe file under .agentic/agents/ in the target repo when you run them (see each entry's why). verdict is "passed" only when every native check passed AND host_must_run is empty; it is "incomplete", never "passed", while host_must_run still has entries — which is every install this server finds, since host_must_run is never empty on a single server-side call. This server alone therefore never returns "passed"; that verdict is only reachable once the host has run the returned commands and folded the result back in. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| agentic-init | Install the agentic-os process layer into the current repo — interview (role presets, HITL dial, autonomy matrix, stack confirm, adapters), dependency registration, template scaffold, stack-specific agent generation with instruction-quality audit, then doctor verification. Journaled and resumable. Use when the user says "/agentic-init", "install agentic-os", "scaffold the agent architecture", "set up the agentic process layer", or "add agentic-os to this repo". |
| agentic-doctor | Standalone verifier for an agentic-os install — checks the file manifest against the install journal, py-compiles AND imports every hook it owns (a badly-rendered scalar compiles but raises on load), dry-runs the enforcement hooks with canned events (block hooks must exit 2 on a synthetic violation, 0 on clean), runs the HITL smoke test on the output-contract gate, verifies settings registration, git hook installation, dependency plugins, scorecard thresholds, and agent-registry table integrity. Writes .agentic/agentic-os/doctor.json. Use when the user says "/agentic-doctor", "verify the agentic-os install", "check the agent setup", "run doctor", or after /agentic-init and /agentic-upgrade. |
| agentic-upgrade | Upgrade a repo's scaffolded agentic-os layer to the currently installed plugin version — per-file three-way reconciliation via the install journal's recorded sha256 (unchanged managed files overwritten, user-modified files diffed and prompted, managed blocks replaced wholesale, generated agents offered regeneration plus re-audit — never a silent overwrite), then doctor re-run and journal stamp bump. Use when the user says "/agentic-upgrade", "upgrade agentic-os", "update the scaffolded agents/hooks", or after updating the agentic-os plugin. |
| sdlc-start | Starts a human-in-the-loop agentic-sdlc run from a task description, an external work-item reference, or a greenfield idea. Triggers include "start sdlc", "implement this with sdlc", "begin SDLC workflow", or a request for the legacy sdlc:start command on a skills-based host. |
| sdlc-task | Runs a lightweight, human-in-the-loop SDLC flow for a task the user has already sized as XS/S/M: brainstorm-lite, spec, plan, inline TDD, one code-review round, and qa-gates on the current feature branch — no complexity scoring, no per-task subagents, no evidence files, no worktrees. Invoke on phrases like "run sdlc-task", "small task, just spec and build it", "quick fix workflow", "this is an XS/S/M task", "lightweight SDLC for this change", or a request to reconcile a finished task's spec/plan against changes made after handoff ("sdlc-task sync", "sync my task", "reconcile the spec with what shipped", mode: sync). Do not invoke for ambiguous or ticket-sized work that needs a full audit trail (use sdlc-start), low-touch full-pipeline automation (use sdlc-autonomous), or simple, research-first work with no spec (use sdlc-light). |
| qe-blueprint-scaffold | Invoke when the user wants to start, scaffold, bootstrap, or implement a quality-engineering blueprint or an agentic test-automation setup — interviews the user, matches intent to one of 28 bundled QE blueprints, and generates a fill-in-ready agent framework (context file, agent stubs, skill stubs) for Claude Code, Cursor, or GitHub Copilot. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| agentic-os/agentic-doctor | Standalone verifier for an agentic-os install — checks the file manifest against the install journal, py-compiles AND imports every hook it owns (a badly-rendered scalar compiles but raises on load), dry-runs the enforcement hooks with canned events (block hooks must exit 2 on a synthetic violation, 0 on clean), runs the HITL smoke test on the output-contract gate, verifies settings registration, git hook installation, dependency plugins, scorecard thresholds, and agent-registry table integrity. Writes .agentic/agentic-os/doctor.json. Use when the user says "/agentic-doctor", "verify the agentic-os install", "check the agent setup", "run doctor", or after /agentic-init and /agentic-upgrade. |
| agentic-os/agentic-init | Install the agentic-os process layer into the current repo — interview (role presets, HITL dial, autonomy matrix, stack confirm, adapters), dependency registration, template scaffold, stack-specific agent generation with instruction-quality audit, then doctor verification. Journaled and resumable. Use when the user says "/agentic-init", "install agentic-os", "scaffold the agent architecture", "set up the agentic process layer", or "add agentic-os to this repo". |
| agentic-os/agentic-upgrade | Upgrade a repo's scaffolded agentic-os layer to the currently installed plugin version — per-file three-way reconciliation via the install journal's recorded sha256 (unchanged managed files overwritten, user-modified files diffed and prompted, managed blocks replaced wholesale, generated agents offered regeneration plus re-audit — never a silent overwrite), then doctor re-run and journal stamp bump. Use when the user says "/agentic-upgrade", "upgrade agentic-os", "update the scaffolded agents/hooks", or after updating the agentic-os plugin. |
| agentic-qe/eval-harness | Scaffold a two-layer eval framework — deterministic contract checks plus optional LLM-judge behavioral cases — for every Claude Code skill (and agent file) in a repository. Invoke when the user asks to add evals, regression guards, or skill tests, to wire eval automation into hooks/CI, or to extend an existing evals.json setup. Not for writing ordinary unit tests or for merely running an eval suite that already exists. |
| agentic-qe/qe-blueprints | Invoke when the user wants to start, scaffold, bootstrap, or implement a quality-engineering blueprint or an agentic test-automation setup — interviews the user, matches intent to one of 28 bundled QE blueprints, and generates a fill-in-ready agent framework (context file, agent stubs, skill stubs) for Claude Code, Cursor, or GitHub Copilot. |
| agentic-sdlc/code-review-orchestrator | Resolves the SDLC code-review gate with a multi-lens review fan-out and persists one canonical verdict JSON as the hand-off back to decision-router. Invoked inline (Skill tool) by decision-router at gate ids `code-review.final` (full multi-lens pass after implementation evidence) and `code-review.check` (narrow re-check of prior blocking findings against a fix-up diff). Trigger when the router reaches "code review", "resolve the review gate", "run code review", "re-check the prior findings", or "verify the fix-up diff". Not for direct user discovery — the router owns invocation. |
| agentic-sdlc/code-review | Standalone, on-demand multi-lens code review of local git changes, outside any managed SDLC pipeline. Invoke when the user says "code review", "review my changes", "review my code", "review my branch", "review staged changes", "review this diff", or "run a code review" — and no pipeline is already driving the review. Resolves a review scope, materializes a diff, hands it to the review orchestrator, and renders the machine verdict. Read-only (R0): reports findings, never edits source, never applies fixes, never commits. |
| agentic-sdlc/complexity-scoring | Thin gating skill that classifies a task's complexity when the caller's own heuristics cannot confidently route it. Invoke when you need to decide between going straight to detailed planning versus running a design/ideation pass first, or when a task might be too large to plan directly. Trigger phrases include "score this task's complexity," "run the complexity gate," "decide writing-plans vs brainstorming," "is this task big enough to need a design phase," and "check if this needs to be split up." Called by sdlc-pipeline as a standing step, and by sdlc-task or sdlc-light only as a fallback when their lightweight heuristics can't decide. Delegates the actual scoring to the sizing-analyst agent and normalizes its output into a fixed {score, routing, breakdown} schema. |
| agentic-sdlc/decision-router | Invoke this skill whenever an SDLC pipeline phase reaches a judgment gate and needs a resolved, logged verdict — spec approval, plan approval, QA drift, code review (final or check round), requirements ambiguity, spec clarification, or feature verification. Trigger phrases include "resolve this gate", "route this decision", "get a verdict for <gate_id>", "who approves this", "check the escalation rule", or "record this to decisions.jsonl". It is the single entry point for every recognized gate id in both hitl and autonomous modes. Do not invoke it for the `qa.ready` signal — that stays deterministic inside the calling pipeline and never enters this router's table. |
| agentic-sdlc/feature-verification | Required functional proof for anything user-visible, run right after qa-gates. When a change touches UI or another externally visible surface, it reuses e2e coverage that already exists, or generates a focused Playwright check when none does; either way it captures screenshots, console output, and network errors, and writes verification-evidence.json per feature. |
| agentic-sdlc/mr-creator | Commit with a mandatory ticket reference, push, and open an MR or PR. Commit, branch, and title conventions are read from `.agentic/guides/standards/git-workflow.md`; the MR/PR adapter (a CLI, MCP server, or custom command) is read from the `## MR Adapter` section of `.agentic/guides/project.md`. No source-control platform is hardcoded. Invoke on explicit review-workflow requests — "commit changes", "push changes", "create MR", "make merge request", "open a PR", and similar. Every commit is rejected unless it carries a ticket reference in the guide's pattern. |
| agentic-sdlc/mr-watch | Watch one open MR/PR and autonomously clear whatever blocks merge — CI failures, requested-change reviewer comments, and merge conflicts — on a polling loop until the request merges, closes, or the user stops it. Adapter-driven: no source-control platform is hardcoded. Invoke when the user says "watch the MR", "monitor MR", "keep an eye on the PR", "babysit this PR", "monitor MR !123", or when an MR-creator skill hands off a freshly opened request for hands-off management. Blast radius R3. |
| agentic-sdlc/product-owner | Turn a raw feature idea, improvement, or bug description into a structured user story with testable acceptance criteria, persisted as a local markdown file. Invoke when the user wants to create, draft, or refine a story, ticket, requirement, or acceptance criteria — trigger phrases include "create a story", "draft a story", "write a user story", "story for this feature", "I need a ticket", "write acceptance criteria", "act as product owner", "spec out this idea", "define requirements", "break this into user stories", or "write an FRD". Fire even when the word "story" is never said: any request to shape a feature, improvement, or fix into a written, agreed requirement qualifies. Works standalone in any repository with no prerequisite project files; the skill explores the codebase for context before it asks the user anything. |
| agentic-sdlc/qa-case-generator | Turn one work-item id into reviewed, backend-synced functional test cases. Invoke when the user says "generate test cases for PROJ-123", "write test cases from this ticket", "qa cases for <ticket>", or hands you a single Jira / Azure DevOps / GitHub / GitLab work-item id and wants manual or API-level functional cases. Produces case documents under docs/superpowers/qa-tasks/, gates them behind a human review, and optionally pushes to a test-management adapter. Not for unit/integration tests, code-coverage, or automation setup — and never auto-invoked by other pipeline skills. |
| agentic-sdlc/qa-e2e-generator | Turn one work-item id into executable end-to-end (UI/API) test automation. Invoke when a QA engineer hands you a ticket id and wants a full E2E suite generated, run, reviewed, and pushed for review. Triggers: "generate e2e tests for PROJ-123", "qa-e2e-generator PROJ-123", "write end-to-end tests for this ticket", "automate e2e for <ticket>", "build a Playwright suite from this story". Standalone orchestrator — NOT auto-invoked by sdlc-pipeline. Requires a prior QA foundation (qa-strategy.md) and a configured ticket adapter (project.md). Top-of-pyramid only: no unit tests. |
| agentic-sdlc/qa-foundation | Bootstrap the QA knowledge foundation for a repo: discover test files, coverage reports, CI test gates, and QA docs; configure adapters to external QA sources (test-case management, QA documentation, external test repos); then write two persistent guides — qa-strategy.md and qa-health.md — under .agentic/guides/testing/. Invoke on first-time QA setup, when the user says "qa init", "qa foundation", or "set up qa", or via the slash command /sdlc:qa-init. These guides are consumed downstream by the qa-planner skill on every feature run, so run this once before the first /sdlc:start or /sdlc:story. |
| agentic-sdlc/qa-gates | Run the host project's quality gates in a fixed order (lint -> build -> unit tests -> affected tests -> optional UI tests) and return a structured, machine-readable merge-readiness report. Runner-agnostic: detects the build tool (npm/pnpm/yarn/cargo/poetry/uv/go) from manifest files, caches a gate plan, and defers browser-level proof to feature-verification. Invoke during a change's implementation phase when an orchestrator needs to know whether the diff is mechanically merge-ready — triggered by "run qa gates", "run the quality gates", "check merge readiness", "run lint/build/tests on this branch", or the SDLC pipeline's post-implementation QA step. This skill reports readiness only; it never creates an MR/PR. |
| agentic-sdlc/qa-planner | Plans and reviews QA on a per-feature basis, in three modes that sdlc-pipeline invokes directly: --checklist (Phase 6) turns requirements and QA knowledge into qa-checklist.md before implementation starts; --review-tests (Phase 8) reviews the tests that were written for quality and completeness; --update (Phase 11) refreshes qa-health.md once qa-gates has passed. Not meant to be invoked directly by a user. |
| agentic-sdlc/release-manager | Audit a software release by cross-referencing git commits in a release range against tickets carrying the release's fix version, then emit a gap-analysis report in Markdown tables. Invoke when the user says "validate the release", "release audit", "check what shipped in <version>", "cross-reference commits and tickets", "did every ticket get committed", "pre-release check", or asks which fix-version tickets have no matching commits (and vice versa). Read-only (R0) — reports mismatches, never mutates tags, tickets, or fix versions. |
| agentic-sdlc/repo-audit-guides | Read-only audit of a repository's documentation, structure, and agentic assistant setup, run before knowledge planting so the downstream repo-guides workflow knows how to handle every pre-existing doc and assistant surface. Invoke when the user asks to survey, audit, inspect, assess, or review repo docs, assistant instructions, Claude / Codex / Gemini / GitHub Copilot setup, subagents, skills, hooks, AGENTS.md, CLAUDE.md, GEMINI.md, or whether a repository is ready for knowledge planting. Produces a structured readiness report, never a numeric grade, and never writes. |
| agentic-sdlc/repo-guides | First-time knowledge setup for a repository: survey the stack, generate the curated guide tree under .agentic/guides/, and wire the AI entrypoint (AGENTS.md plus a CLAUDE.md shim). User-triggered only. Invoke when the user says "repo guides", "set up repo guides", "generate the guides", "bootstrap this repo for AI", "agentic init", "create AGENTS.md", or "wire the AI entrypoint". On a brownfield repo, run repo-audit-guides first and feed its "# Repo Knowledge Audit" in here. Writes tracked files (blast radius R2) — no writes happen until the user approves the plan. |
| agentic-sdlc/requirements-intake | Normalize a single starting input — free-form text, an external ticket id or URL, a local story path, a local work-item path, or a greenfield idea — into one canonical requirements.md that every later pipeline phase reads, and guarantee a work-item reference exists before those requirements are written. Invoked automatically by sdlc-pipeline at Phase 1; also triggers on direct asks like "normalize these requirements", "turn this into requirements", or "write requirements from this ticket". Only normalizes and clarifies — never plans, estimates, scaffolds, or implements. |
| agentic-sdlc/role-memory | Per-role persistent memory for agents — durable curated facts, preferences, and decisions plus an append-only episodic daily log, stored as plain markdown under `.agents/memory/<role>/`. Invoke when the user says "remember this", says "log this", or asks "what did you learn yesterday" — and whenever you observe a fact, correction, preference, or decision worth carrying across sessions, or need to recall what you already know before acting. Uses Read/Write/Edit/Glob only: no CLI, no scripts, host-independent (Claude Code, Cursor, Gemini CLI, Windsurf, Copilot CLI, or an external supervisor). |
| agentic-sdlc/sdlc-autonomous | Start an autonomous ("factory mode") SDLC run from a task description, external work-item reference, story path, or greenfield idea, on hosts that support skills but not custom slash commands. Trigger on "run autonomously", "factory mode", "ship this without asking", or a request for the legacy `sdlc:autonomous` command. Parses free-form intent into a structured payload and hands off to sdlc-pipeline — carries no orchestration logic of its own. |
| agentic-sdlc/sdlc-doctor | Force-runs a full environment-readiness check for the agentic-sdlc plugin and overwrites .agentic/agentic-sdlc/doctor.json with the fresh result, ignoring any existing fingerprint or TTL cache. Invoke when the user asks to verify the agentic-sdlc setup, re-run doctor, check whether superpowers/node/git are ready, or asks for the legacy sdlc:doctor command on a host (for example, a Codex-class host) that has no native slash-command support and must substitute a skill for it. |
| agentic-sdlc/sdlc-light | Lightest-ceremony entry point into the agentic-sdlc pipeline for a task the user has already judged small and unambiguous. Invoke on phrases like "just do this quickly", "small fix, skip the ceremony", "simple task, no spec needed", or "run sdlc-light on <task>". Runs a mandatory research pass via the codebase-scout agent, a binary Clear/Unclear clarity check capped at 3 questions instead of complexity scoring or brainstorming, skips spec.md entirely, then proceeds through planning, QA checklist, inline TDD, QA test review, up to two code-review rounds, quality gates, and a QA health update, ending in a handoff message rather than an automatic merge or PR. Do not invoke for ambiguous or ticket-sized work (use sdlc-start or sdlc-autonomous) or when the user has already classified task size XS/S/M (use sdlc-task instead). |
| agentic-sdlc/sdlc-pipeline | Heavy orchestrator that runs the full 13-stage (Phase 0-12) governed SDLC flow for a single unit of work. Invoke this skill only as the execution engine dispatched by sdlc-start (mode=hitl) or sdlc-autonomous (mode=autonomous) — never in direct response to a bare user request, and never with a mode other than hitl or autonomous. Trigger phrases that mean "call sdlc-pipeline" arrive pre-translated through those two entry skills: "start sdlc", "implement this with sdlc", "run autonomously", "factory mode", "ship this without asking", a resumed run handed back from sdlc-status. Owns run bootstrap, branch safety, complexity routing, spec/plan approval, TDD implementation with evidence capture, deferred two-round code review, QA gates, feature verification, and handoff. mode is the only branching input between the two calling styles: phase sequence, artifact shapes, file paths, and gate ids are identical either way — only the decision-router behavior at each judgment gate differs. |
| agentic-sdlc/sdlc-start | Starts a human-in-the-loop agentic-sdlc run from a task description, an external work-item reference, or a greenfield idea. Triggers include "start sdlc", "implement this with sdlc", "begin SDLC workflow", or a request for the legacy sdlc:start command on a skills-based host. |
| agentic-sdlc/sdlc-status | Read-only inspector and resumer for agentic-sdlc heavy-pipeline runs kept under docs/superpowers/runs/. Invoke when the user asks what a run's status is, wants to list or drill into past pipeline runs, needs to pick back up an interrupted run, or is on a skill-only host (for example, Codex) that has no `sdlc:status` slash command and needs the same behavior exposed as a skill. Trigger phrases: "sdlc status", "what's the run status", "list sdlc runs", "resume the pipeline", "resume run <id>", "is the run stuck", "sdlc:status". Never mutates a completed run, and never resumes an aborted one without an explicit yes. |
| agentic-sdlc/sdlc-task | Runs a lightweight, human-in-the-loop SDLC flow for a task the user has already sized as XS/S/M: brainstorm-lite, spec, plan, inline TDD, one code-review round, and qa-gates on the current feature branch — no complexity scoring, no per-task subagents, no evidence files, no worktrees. Invoke on phrases like "run sdlc-task", "small task, just spec and build it", "quick fix workflow", "this is an XS/S/M task", "lightweight SDLC for this change", or a request to reconcile a finished task's spec/plan against changes made after handoff ("sdlc-task sync", "sync my task", "reconcile the spec with what shipped", mode: sync). Do not invoke for ambiguous or ticket-sized work that needs a full audit trail (use sdlc-start), low-touch full-pipeline automation (use sdlc-autonomous), or simple, research-first work with no spec (use sdlc-light). |
| agentic-sdlc/test-heal | Repairs failing tests whose failure is the test's own fault — and only those. Consumes a failure triage (test_issue | environment_issue | flaky | application_issue), fixes test code, fixtures, selectors, and waits, never application code, and never re-runs the suite itself: it applies fixes, sanity-checks that they parse, commits once, and hands a machine-readable Loop Decision back to the orchestrator, which owns re-running qa-gates. application_issue failures are returned untouched as pipeline fix-up work. |
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/Jarroslav/agentic-os'
If you have feedback or need assistance with the MCP directory API, please join our Discord server