Instalação portátil:
docs/installation/quickstart.md— clone → setup → install. Delegação agêntica: forge.agentic.json (SpecialistAgenticManifest/v1). Docs: commands · skills · agents · tutorial · economy · troubleshooting
Platform Forge
Agentic Platform Engineering intelligence platform — deterministic, offline-first, evidence-first, graph-aware, provider-neutral in the core.
Platform Forge reads repositories, Terraform/OpenTofu, Kubernetes, GitOps, CI/CD, observability data, SLOs, security posture, cost and ownership — and builds one coherent, evidence-backed model of a platform. Facts have provenance; findings cite evidence; unanswered questions come back as named refusals, never hedged prose.
Versão em Português: README.pt-BR.md
Install
pip install -e . # core — offline-capable
pip install -e ".[dev]" # + tests/lint
# or: uv sync && uv pip install -e .
Quick start
platformforge doctor # environment health
platformforge init --repo . # scaffold .platformforge/
platformforge inspect --repo . # inventory analyzable artifacts
platformforge analyze iac ./infra # HCL → facts
platformforge analyze k8s ./manifests # manifests → facts
platformforge analyze gha --repo . # workflows → facts
platformforge analyze kyverno ./policies # kyverno/gatekeeper → facts (version-aware)
platformforge analyze cosign bundle.json # sigstore shape — claimed ≠ verified
platformforge analyze slsa prov.json # SLSA requirement/evidence/gap
platformforge analyze cloud-aws ./dumps # AWS CLI dumps → T1 facts
platformforge judge facts.json # rules → findings
platformforge judge facts.json --versions '{"kubernetes":"1.29"}' # version-gated rules
platformforge graph build facts.json # facts → provenanced graph
platformforge graph blast --node X # blast radius (per impact class)
platformforge graph identity-become --node role/x # who can become this role
platformforge observe slo contract.yaml --sli '{"total_events":1e6,"bad_events":400}'
platformforge observe postmortem --incident i.json # unresolved root cause stays named
platformforge finops costs billing.json # cost facts + summary
platformforge finops ingest cur.csv # CUR/Azure/GCP/OpenCost/Kubecost → rows
platformforge finops focus-validate rows.json # FOCUS 1.0 column compliance
platformforge finops unit costs.json --denominators d.json # cost/request etc.
platformforge change review --repo . --patch diff.patch # sandbox → graph delta → risk
platformforge product maturity --signals s.json
platformforge mcp tools # bounded MCP capability surface
platformforge mcp serve # stdio server (same core as CLI)
platformforge lab run-all # Forge Lab scenario suite
platformforge lab chaos lab/chaos-pod-kill # fault injection on the graph
platformforge evals run # eval framework (§94–95)
platformforge evals coverage # rule ↔ case coverage matrix
platformforge evals precision # measured false-positive rate
platformforge store gc # §151 — dry-run GC of stale blobs
platformforge collect ./dumps # sniff dumps → facts
platformforge fleet report lab/fleets/acme # fleet north-star receipt
platformforge optimize scan lab/fleets/acme # opportunities → ChangeIntent
platformforge agents list # canonical 41-agent roster
platformforge agents check # host-mirror drift gate
platformforge agents playbook "task" # zero-subagent fallback
platformforge route '{"task":"...","risk":"high"}' # Router V2 decision
platformforge bench scale # measured 10k-node/500k-edge bench
platformforge diagnose <node> --findings j.json --facts f.json
platformforge plan findings.json # ordered remediation plan
platformforge policy check facts.json # catalog as policy
platformforge security ./dumps # secrets+iam+sbom+supply bundle
platformforge sdd status --feature F # spec lifecycle + hash cascade
platformforge forge manifest # capability manifest (interop)
Pipeline
artifacts → analyze → facts ──► judge → findings ──► scorecards / gates
│ ▲ rules/catalog/
└──────► graph build → graphfy ──────┘ (deps, blast, diff)
Facts carry evidence tiers (measured → provider → plan → repo → doc →
declared → inferred). Graph edges inherit provenance. Every mutation path
goes through inspect → propose → sandbox → verify → approve → apply;
the core itself never mutates anything.
Design docs
Index: docs/README.md — ADRs, agent docs, cycle reports.
Architecture & model: ARCHITECTURE.md · GRAPHFY.md · DOMAIN-MAP.md · RUNTIME-TOPOLOGY.md · RECONCILIATION.md · CLOUD.md · OBSERVATION-MODEL.md · COMPATIBILITY.md · DEPRECATION.md
Scale & ops: FLEET.md · MULTI-CLUSTER.md · PLATFORM-ANALYTICS.md · OPTIMIZATION.md · FEDERATION.md · OPS.md · APPROVALS.md · ROLLBACK.md · LIVE.md · COLLECTORS.md · CAPACITY.md · AI-PLATFORM.md · ENTERPRISE.md · GOLDEN-PATHS.md · GOLDEN-PATH-ANALYTICS.md · POLICY-INTELLIGENCE.md · PLATFORM-MEASUREMENT.md · INCIDENTS.md · RUNBOOKS.md · EXECUTION.md
Agentic & economy: AGENTIC-OS.md · docs/agents/ (agentic runtime) · ECONOMY.md · QUALITY-PER-TOKEN.md · KNOWLEDGE.md · SDD.md · VERIFICATION.md
Rules & evidence: RULES.md · EVALS.md · SECURITY.md · SOURCES.md · CAPABILITIES.md · RESEARCH.md · docs/freeze/ (architecture freeze)
Reference: CLI-REFERENCE.md (curated) · docs/CLI-SURFACE.md (generated index) · GLOSSARY.md · MCP.md · ROADMAP.md · CHANGELOG.md · CONTRIBUTING.md · AGENTS.md · CODE_OF_CONDUCT.md · README.pt-BR.md
Support matrix
| Axis | Supported |
|---|---|
| Python | 3.10+ (CI tests 3.10–3.13) |
| OS | Linux (primary); macOS/Windows run the core — host adapters are platform-dependent |
| Runtime deps | PyYAML, jsonschema, python-hcl2 — offline core only |
| Optional extras | aws (read-only dumps), mcp (host surface), dev (pytest, ruff) |
| Host agent mirrors | .claude/, .agents/, .codex/, .devin/ — generated, agents check gated |
| Architecture state | Frozen — see docs/freeze/; new capability requires an unfreeze RFC |
License
MIT — see LICENSE, SOURCES.md, CREDITS.md.
Graph Studio
platformforge graph view|ui projects the built graph through
forge/ForgeGraphView/v1 and serves the embedded local explorer (declinable at
install via --components). Full docs: the-forge/docs/graph-studio/.
Metadata
Release files for platformforge 0.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| platformforge-0.1.0.tar.gz | 58.4 MB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| platformforge-0.1.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 59.5 MB
Release files / platformforge-0.1.0.tar.gz
| Download URL | platformforge-0.1.0.tar.gz |
|---|---|
| Size | 58.4 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
941d469714518f868ad3e51a362fa14979f15f25aefe74b606776a5be86adf48
|
|
BLAKE2b-256 checksum How to use checksums |
7e00eb26f723b12289954f70ad49fcd1cc5c3d1f4b97f78f33872988c4575203
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.15
|
Release files / platformforge-0.1.0-py3-none-any.whl
| Download URL | platformforge-0.1.0-py3-none-any.whl |
|---|---|
| Size | 1.1 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
79f601bcf3fb5b546846d61ae4dfcc03538e28da9364ec18b599f4e9efdde0ab
|
|
BLAKE2b-256 checksum How to use checksums |
d28e7c540d075f7fe5c0137ec34055f953128c17f08e112dec066f1b2906c9fc
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.15
|