Skip to content

Releases

FORGE v2.0.0

Choose a tag to compare

@github-actions github-actions released this 22 Nov 11:17
88e9aa6

FORGE Changelog

2.0.0 — The verified agent harness

FORGE 2.0 turns the workbench into a harness layer for the AI coding tools you already use — and keeps FORGE's evidence-first approach: install, guard, verify and prove.

Install into every harness

  • forge install renders FORGE agents, skills, slash commands and coding rules into Claude Code, Cursor, Codex, OpenCode and Gemini native formats, with minimal, standard and full profiles and automatic language detection (TypeScript/Node, React, Python, Go, Rust, Java/Kotlin, C#, PHP, Ruby, Swift).
  • Every written file is hashed in .forge/install-state.json; FORGE never overwrites files it did not write or that you edited (unless --force), only edits its own managed block in AGENTS.md/GEMINI.md, and refuses to write through symlinks.
  • forge doctor, forge repair, forge uninstall (restores the original project) and forge catalog.
  • Catalog: 41 agents, 26 skills, 24 commands and 23 rules, each validated in CI.

FORGE Shield

  • forge shield scores agent configuration 0–100: Claude Code permissions and hooks, MCP servers (Claude, Cursor, VS Code, Gemini, OpenCode), hidden prompt injection (invisible Unicode, override phrases, hidden comments), secrets and CI workflow risks.
  • JSON, Markdown and SARIF 2.1.0 output, --fail-on CI gate, safe --fix; defensive guidance that quotes attack phrases is not flagged.
  • GitHub Action operation: shield; FORGE's own CI self-scans.

Runtime hooks and memory

  • forge install --hooks adds Claude Code hooks: FORGE Guard (denies catastrophic commands, asks before force-push, history discard, destructive SQL, publish/deploy, sudo, secret access and agent-config edits), a Shield check when agents edit their own configuration, an optional verification gate on Stop, and session-start memory injection.
  • Layered memory (project + global) with provenance, credential rejection, relevance + recency search, handoffs; MCP tools forge_memory_search, forge_memory_add, forge_shield_scan.

GitHub App and hosted dashboard

  • Comment /forge analyze to get a setup pull request with FORGE files and a generated repository-specific forge-project skill (real verify commands, conventions, commit style). Also /forge repair, audit, doctor, risk, help.
  • Every pull request gets a FORGE check run with an explainable change-risk score and Shield annotations.
  • Hosted dashboard at /app/: Shield scores, setup health, PR risk, stack, activity and one-click setup PRs for repositories you can access.
  • forge analyze --write generates the project skill locally.

Upgrade notes

  • The GitHub App needs Contents, Pull requests, Issues and Checks read & write, plus Issue comment and Pull request events. Sign in to the dashboard again after upgrading.
  • forge memory context --query no longer mistakes the query for a project path.
  • All 1.0 commands keep working unchanged.

1.0.0 — First major release

FORGE 1.0.0 brings the independent FORGE engineering workbench together as a versioned, installable product. It is designed to provide a local-first alternative for teams and developers who want codebase intelligence, agent orchestration, reviewable workflows, and measurable verification.

Product experience

  • Local engineering dashboard with project overview, health snapshots, architecture and dependency graphs, run history, task inspection, approvals, traces, reviews, verification history, and benchmark/evaluation views.
  • Workspace explorer and code search for inspecting project files and symbols.
  • Agent Configuration Studio for custom agent and skill Markdown, snapshots, and benchmark-gated configuration promotion.
  • Visual workflow editor for reusable specialist-agent pipelines and draft run creation.
  • Provider readiness and diagnostics views that avoid exposing credential values.

Engineering capabilities

  • Project inspection, architecture mapping, dependency intelligence, engineering-risk analysis, change intelligence, and quality auditing.
  • Specialist-agent orchestration with saved runs, incremental task settlement, approval gates, replay/retry, and resume recovery.
  • Project memory, workspace skills, declarative hooks, permission policies, and bounded read-only workspace tools.
  • Review history, dry-run patch validation, explicit patch approval records, and staged promotion APIs.
  • Verification history, evaluation suites, benchmark matrices, regression comparisons, and experiment manifests.
  • MCP server and CLI workflows for local engineering automation.

Install and launch

npm install -g forge-engineering-workbench
forge --version
forge init ./my-project
forge dashboard ./my-project

The dashboard binds to localhost by default. Open the URL printed in the terminal.

Important V1 boundaries

  • The public Vercel website is a static product site. The live dashboard runs locally against a workspace selected on the user's machine; it is not a hosted multi-user dashboard.
  • Provider-backed runs require the user's own supported provider configuration and credentials. No provider connection or API spend is included by this release.
  • Native model tool-calling is currently limited to bounded read-only workspace tools for the OpenAI-compatible provider. Other documented providers remain proposal-only in the normal run path.
  • MCP workspace tools are read-only. Patch validation is a dry run; patch promotion is a separately reviewed library API, not an automatic agent action.
  • Local script verification executes project-controlled code on the host. Docker verification is opt-in and depends on the host runtime. Do not treat local execution or heuristic risk scores as a security guarantee.
  • Rootless Docker, live provider end-to-end validation, and hosted multi-user operation are not claimed as verified V1 capabilities.

This release establishes the FORGE product foundation. Future releases can expand integrations and hosted operation without weakening the local-first permission and approval model.

What's Changed

  • docs: refresh FORGE README and product guide by @ansh44233 in #14
  • Fix website integration status for Claude Code and Cursor by @ansh44233 in #15
  • Rebuild README with FORGE WORKBENCH identity and full project guide by @ansh44233 in #16
  • feat: add GitHub App installation webhook foundation by @ansh44233 in #17
  • feat: add secure GitHub App OAuth sessions by @ansh44233 in #18
  • fix: resolve Vercel API TypeScript import errors by @ansh44233 in #19
  • fix: import Redis helper from GitHub App module by @ansh44233 in #20
  • feat: add GitHub account connection controls by @ansh44233 in #21
  • fix: point header GitHub link to account profile by @ansh44233 in #22
  • feat: simplify GitHub App installation with Get App by @ansh44233 in #23
  • docs: refresh FORGE README and simplify Get App entry point by @ansh44233 in #24
  • docs: redesign README with ECC-style project presentation by @ansh44233 in #25
  • docs: redesign README branding and visual hierarchy by @ansh44233 in #26
  • docs: make FORGE AI harness identity and README visuals clear by @ansh44233 in #27
  • docs: fix broken GitHub Trending and Star History rank badges by @ansh44233 in #28
  • docs: add real logos to compact README badges by @ansh44233 in #29
  • docs: enlarge and align README logo marks by @ansh44233 in #30
  • docs: add FORGE support contact by @ansh44233 in #31

Full Changelog: v1.0.0...v2.0.0

FORGE v1.0.0

Choose a tag to compare

@ansh44233 ansh44233 released this 17 Nov 01:29
e2e8193

FORGE Changelog

1.0.0 — First major release

FORGE 1.0.0 brings the independent FORGE engineering workbench together as a versioned, installable product. It is designed to provide a local-first alternative for teams and developers who want codebase intelligence, agent orchestration, reviewable workflows, and measurable verification.

Product experience

  • Local engineering dashboard with project overview, health snapshots, architecture and dependency graphs, run history, task inspection, approvals, traces, reviews, verification history, and benchmark/evaluation views.
  • Workspace explorer and code search for inspecting project files and symbols.
  • Agent Configuration Studio for custom agent and skill Markdown, snapshots, and benchmark-gated configuration promotion.
  • Visual workflow editor for reusable specialist-agent pipelines and draft run creation.
  • Provider readiness and diagnostics views that avoid exposing credential values.

Engineering capabilities

  • Project inspection, architecture mapping, dependency intelligence, engineering-risk analysis, change intelligence, and quality auditing.
  • Specialist-agent orchestration with saved runs, incremental task settlement, approval gates, replay/retry, and resume recovery.
  • Project memory, workspace skills, declarative hooks, permission policies, and bounded read-only workspace tools.
  • Review history, dry-run patch validation, explicit patch approval records, and staged promotion APIs.
  • Verification history, evaluation suites, benchmark matrices, regression comparisons, and experiment manifests.
  • MCP server and CLI workflows for local engineering automation.

Install and launch

npm install -g forge-engineering-workbench
forge --version
forge init ./my-project
forge dashboard ./my-project

The dashboard binds to localhost by default. Open the URL printed in the terminal.

Important V1 boundaries

  • The public Vercel website is a static product site. The live dashboard runs locally against a workspace selected on the user's machine; it is not a hosted multi-user dashboard.
  • Provider-backed runs require the user's own supported provider configuration and credentials. No provider connection or API spend is included by this release.
  • Native model tool-calling is currently limited to bounded read-only workspace tools for the OpenAI-compatible provider. Other documented providers remain proposal-only in the normal run path.
  • MCP workspace tools are read-only. Patch validation is a dry run; patch promotion is a separately reviewed library API, not an automatic agent action.
  • Local script verification executes project-controlled code on the host. Docker verification is opt-in and depends on the host runtime. Do not treat local execution or heuristic risk scores as a security guarantee.
  • Rootless Docker, live provider end-to-end validation, and hosted multi-user operation are not claimed as verified V1 capabilities.

This release establishes the FORGE product foundation. Future releases can expand integrations and hosted operation without weakening the local-first permission and approval model.

What's Changed

  • fix(dashboard): harden loopback Host validation by @ansh44233 in #9
  • ci(release): validate release tag against package version by @ansh44233 in #10
  • feat(security): integrate agent configuration audit by @ansh44233 in #12
  • chore(action): prepare Marketplace metadata by @ansh44233 in #13

Full Changelog: https://github.com/ansh44233/FORGE-ECC/commits/v1.0.0