{
"markdown": "![\"Agentic](\"docs/logo.svg\")
\n\nAn open profiler for agentic coding methodologies, built to help you find the ones that fit you and your projects.\n\nAn \"agentic methodology\" is the plugin, framework, or skill collection you add to your coding harness (Claude Code, Cursor, Codex, Gemini CLI, etc). Popular examples include [Superpowers](https://github.com/obra/superpowers), [BMAD-METHOD](https://github.com/bmad-code-org/BMAD-METHOD), [GSD](https://github.com/jnuyens/gsd-plugin), [LFG](https://mcpmarket.com/tools/skills/lfg-autonomous-engineering-workflow), but Agentic Atlas can profile any framework, method, or skill collection.\n\nThere's no single best agentic methodology, only a best fit for your project and how you like to work. Agentic Atlas shows you where each tool sits so you can pick.\n\n## What it does\n\nAgentic Atlas reports what it sees, it doesn't grade, rank, or crown a winner, because every person and project has unique needs and collapsing independent axes into one number tells you nothing. It locates each tool on signed, diverging axes: `0.0` is neutral, the sign says which pole the tool leans toward, and the magnitude says how strongly. A tool at `-7.8` on Greenfield ↔ Brownfield isn't worse than one at `+2.0`, it is simply aimed at greenfield projects, and you are shown this to draw your own conclusions on fit.\n\nEach position is a deterministic function of small, named, evidence-backed indicators, every one cited to the target repository. So when you disagree with where a tool landed, trace it to the exact indicators behind it and open a pull request against the versioned rubric. The goal is for the rubric to be a community-maintained, open, and auditable standard for profiling agentic methodologies.\n\nSee [`docs/design.md`](docs/design.md) for the architecture and [`docs/axes.md`](docs/axes.md) for the axis authoring method and the full candidate catalog.\n\n## How an axis is scored\n\nAn axis is a signed spectrum (negative/positive float) between two named poles, with a symmetric scale (default `10`, so scores run `-10` to `+10`). Neither pole is a failure mode.\n\nYou never score an axis directly. It decomposes into indicators, each a narrow question with a bounded answer mapping to a value in `[-1, 1]` signed toward one pole. Two kinds:\n\n- **measured**: computed by the engine from the repository, no model (vocabulary density, file presence, git statistics, GitHub stars). Same input, same output.\n- **classified**: a model reads the repository and picks from a fixed answer set, backing the choice with a quote copied verbatim from the target. The engine rejects any answer whose quote it can't find, so nothing rests on an unverified claim.\n\nThe score is then arithmetic, over resolved indicators, clamped to `[-scale, +scale]`:\n\n```\naxis_score = scale * sum(weight_i * value_i) / sum(weight_i)\n```\n\nUnresolved indicators are excluded and the profile reports coverage, so you never mistake a partial profile for a complete one.\n\n## Axes in v1\n\nA curated subset of the candidate catalog ships today; the full catalog is much larger, see [`docs/axes.md`](docs/axes.md). Heavily correlated axes dilute a profile rather than sharpen it, so the rest are backlog.\n\n| Axis (negative ↔ positive) | Meaning |\n|---|---|\n| **Interrogative ↔ Opinionated** | Elicits and asks vs prescribes a strong default path |\n| **Human-in-loop ↔ Autonomous** | Frequent checkpoints vs unattended autopilot |\n| **Greenfield ↔ Brownfield** | Excels from an idea vs excels inside an existing codebase |\n| **Small-scope ↔ Large-scope** | One task vs the whole delivery lifecycle |\n| **Prototype ↔ Production** | Fast throwaway output vs production hardening |\n| **Solo ↔ Team** | Single developer vs multi-contributor and team-safe |\n| **Generalist ↔ Specialist** | Any domain vs software delivery specifically |\n| **Fresh ↔ Mature** | New and fast-moving vs established and battle-tested |\n| **Spec-light ↔ Spec-driven** | Jumps to code vs plans and specs first |\n| **Test-optional ↔ Test-first** | Testing incidental vs TDD enforced |\n| **Single-agent ↔ Multi-agent** | One conversation vs specialized subagents or personas |\n| **Prescriptive ↔ Composable** | Fixed pipeline vs pick-and-choose parts |\n| **Lightweight ↔ Heavyweight** | Small footprint and little ceremony vs large and elaborate |\n\nAuthoritative definitions live in `rubric/v1/`.\n\n## Usage\n\nEverything runs through the Makefile. Run `make` to list targets.\n\n```bash\nmake setup # create the venv and install with dev deps\nmake check # lint then test (the CI gate)\nmake validate # validate the rubric against the schema\nmake profile TARGET=/path/to/methodology FORMAT=md\n```\n\nThe `agentic-atlas` CLI is available inside the venv:\n\n```bash\nagentic-atlas validate rubric/v1 # check against the schema\nagentic-atlas docs rubric/v1 # regenerate axis README scoring blocks\nagentic-atlas profile /path/to/methodology # measured indicators, deterministic, no key\nagentic-atlas questions /path/to/methodology # emit the classified questions to answer\nagentic-atlas profile /path/to/methodology --answers answers.json # unlock classified indicators from answers\nagentic-atlas compare bmad-method superpowers gsd # (planned) overlay tools on the same axes\n```\n\nThe engine is deterministic and needs no API key. A bare `profile` run resolves the **measured** indicators, the ones the engine computes directly from the repository, and reports how much of each axis that covers. The **classified** indicators, the ones that need the repository read and interpreted, are unlocked by supplying answers: `questions` lists them, an agent answers each with a value from its fixed set and a quote from the target, and `profile --answers` validates every answer (the quote must appear verbatim, the value must be one of the declared options) and scores the ones that pass. The engine never calls a model; it validates.\n\nThe intended answerer is the `run` skill of the `agentic-atlas` plugin. Its host agent is already a capable model with repo access, so it answers the classified questions and feeds them back, no key required. Running the engine raw gives you the deterministic measured axes; running it through the skill unlocks the rest.\n\n## Plugin installation\n\nThe plugin ships three skills. Install from [agentic-marketplace](https://github.com/adamcaviness/agentic-marketplace); pick **one** path per harness.\n\n- `/agentic-atlas:run [path-or-git-url]` profiles a target and unlocks the classified axes (the flow described above).\n- `/agentic-atlas:open-explorer` opens the hosted **Explorer** at , where ready-made profiles of popular methodologies let you browse and find your fit without running anything yourself.\n- `/agentic-atlas:explain [question]` explains the method and helps you read a profile, grounded in these docs and the rubric, upholding the no-ranking, no-aggregate-score stance.\n\n\nClaude Code
\n\n```bash\n/plugin marketplace add adamcaviness/agentic-marketplace\n/plugin install agentic-atlas@agentic-marketplace\n```\n\n \n\n\nCursor
\n\nSee [.cursor/INSTALL.md](.cursor/INSTALL.md). If you already installed via Claude Code, stop there. Cursor-only: clone into `~/.cursor/plugins/local/agentic-atlas`. Teams admins import the marketplace from [cursor.com/dashboard](https://cursor.com/dashboard).\n\n \n\n\nCodex / ChatGPT desktop
\n\nSee [.codex/INSTALL.md](.codex/INSTALL.md). In ChatGPT desktop: **Plugins → Add plugin marketplace**. Source `adamcaviness/agentic-marketplace`, Git ref `main`, Sparse paths empty. Then install **agentic-atlas**.\n\n```bash\ncodex plugin marketplace add adamcaviness/agentic-marketplace --ref main\ncodex plugin add agentic-atlas@agentic-marketplace\n```\n\n \n\n## Reproducibility and fairness\n\nThe rubric (`rubric/`) is versioned data (axes, poles, indicators, weights) and the engine (`agentic_atlas/`) is the code that reads it, each under its own semver. See [`docs/versioning.md`](docs/versioning.md). Every profile stamps the rubric version, engine version, target commit SHA, and, for classified indicators, the source of each supplied answer, so any profile is reproducible and arguable.\n\nI also maintain [agentic-toolkit](https://github.com/adamcaviness/agentic-toolkit), which is itself a target and gets profiled with the same rubric and engine as everything else, no special treatment. I built this tool specifically so I could evaluate my own work and get a picture on where it best fits.\n\n## Status\n\nEarly scaffold, actively developed. Working today: the per-axis rubric with schema validation, the deterministic scoring core, evidence collectors (vocabulary, path presence, git stats, GitHub API), the classified-indicator seam (a `questions` worklist plus a quote-verified `--answers` path, no API key), text/markdown/JSON/HTML reports, the `agentic-atlas docs` generator kept in sync by `make docs-check`, the three plugin skills (`run`, `open-explorer`, `explain`), a committed corpus of 18 profiles under `profiles/`, and the hosted **Explorer** built from that corpus by `scripts/build_site.py` and deployed to GitHub Pages. Next: the `compare` overlay and a wider corpus. See `docs/` and `specs/handoff.md`.\n\n## License\n\nMIT. See `LICENSE`.\n",
"bytes": 9567,
"sha": "29af6ea1d6274094720ca70ae373c22f1974cc804a6b4357703a7febacc82281",
"repo_slug": "adamcaviness/agentic-atlas",
"fonte": "repo",
"truncated": false,
"api": "https://agentalog.com/api/listings/plg_adamcaviness_agentic_atlas_63027839/readme"
}