GStack Port for Codex — Extended

Codex-native versions of Garry Tan’s GStack and GBrain, a Praneet extension layer for responsible product judgment, and a tracked Impeccable design-quality integration.

This repository started as a port of GStack. It now has four distinct layers, tracked separately so their provenance stays clear.

Layer	What it contributes	Current registry
GStack	Planning, design, review, debugging, QA, release, and engineering workflows	56 ports
GBrain	Capture, research, ingestion, daily operations, local memory, and knowledge maintenance	53 ports
Praneet extensions	Responsible design, accessibility, research synthesis, design leadership, market judgment, and outcome learning	7 extensions
Impeccable integration	Project-aware design quality, optional deterministic checks, critique evidence, and external live-mode routing	capability map

What changed from upstream

This is an adapted port. The complete 2026-08-08 reconciliation covers all 108 canonical GStack and GBrain packages and all 109 mapped source lineages through GStack 94993f7 and GBrain 0b47afb. skillify retains both upstream lineages without creating a duplicate installed skill. The GStack registry also retains the local gstack router and checkpoint compatibility skill.

• Codex-native skills: upstream commands were rewritten as focused SKILL.md workflows with Codex-oriented invocation and outputs.
• Explicit runtime boundaries: browser, automation, subagent, permission, and external-tool dependencies carry specific availability labels.
• Local GBrain substrate: ambient memory behavior was adapted into explicit Markdown-backed capture, search, linking, ingestion, synchronization, citation, and health-check tools.
• Natural-language routing: workflow-router chooses a small, useful skill set from an ordinary request, so users can invoke workflows without memorizing the catalog.
• Validation and reconciliation: the canonical inventory, alpha, beta, and complete reconciliation manifests keep provenance, retention, removals, review boundaries, and adoption decisions inspectable.
• Tracked design runtime: selected Impeccable guidance is adapted into a concise Codex bridge. Its Apache-2.0 detector, hooks, and live browser runtime remain optional upstream capabilities.

See the GStack compatibility map, GBrain adaptation memo, Impeccable compatibility map, and runtime compatibility notes for the detailed boundaries.

What is distinctly mine

The Praneet layer changes what the system checks and how it makes recommendations.

• responsible-design-review checks autonomy, consent, dark patterns, fairness, vulnerable users, and data dignity.
• accessibility-review makes accessibility a first-class quality bar across assistive technology, cognitive load, motion, language, and interaction quality.
• research-synthesis grades evidence, checks bias, preserves quote-to-insight traceability, and names what remains unknown.
• design-leadership-review adds a CDO-level lens for quality bars, critique, organizational implications, alignment, and durable decision records.
• startup-memo and market-map combine founder judgment with user impact, ethics, and social consequences.
• outcome-memory compares recommendations with real results so future judgment can improve from measured outcomes.

The router also uses a chief-of-staff selection pattern: consider the skills that could add leverage, then deliberately pare them down to the smallest useful set. That favors judgment over process for its own sake.

Try it

Install the planning and engineering workflow layer with the checked-in installer:

python3 scripts/install_skills.py --bundle core

Install the public Praneet extensions separately when they are useful:

python3 scripts/install_skills.py --bundle extensions

Use --dest PATH to install into a project or test directory. The default is the official Codex user-skill location, ~/.agents/skills. Existing skills are preserved. --force updates only directories previously installed by this package, refuses shared symlinks and local edits, and removes only stale files listed in the prior ownership receipt.

Then ask naturally:

Review this feature for user agency, accessibility, and unintended harm.

Synthesize these interviews, grade the evidence, and separate participant voice from interpretation.

Choose the smallest useful workflow for taking this feature from plan review through QA.

Other useful entry points:

• gstack — route across the engineering and product-development surface
• brain-ops — use local memory before outside research or new conclusions
• office-hours — pressure-test an idea before implementation planning
• design-leadership-review — review whether the organization is making the right design decision
• design-quality — apply project context, design gates, optional Impeccable checks, and evidence-aware fallbacks

For the full coding workflow, see the coding workflow guide.

Mutable sources and privacy

GBrain’s local substrate can synchronize a changing authoritative file, such as a living strategy, identity, or operating document. Generated projections stay local:

python3 scripts/brain_sync_source.py path/to/source.md --title "Source title"

When the source changes, the managed projection replaces its current compiled truth while retaining version, hash, and supersession provenance. An unchanged source is a byte-for-byte no-op. Generated brain content is ignored by Git by default; only the public brain/README.md contract is tracked.

Honest boundaries

• native means the workflow maps cleanly to Codex.
• workflow-adapted means the operating intent is preserved but the interaction or runtime model changed.
• runtime-aware means the skill still depends on available browser, session, automation, device, credential, or host tooling.
• Impeccable external-runtime capabilities stay upstream and run only when already installed in the target project.
• The repository includes a file-backed brain substrate and helper scripts. User brain corpora stay local.
• Registry pins are conservative snapshots. Run the drift checker before claiming current upstream parity.

Repository map

• skills/ — GStack, GBrain, and Praneet workflow definitions
• data/ — separate provenance and compatibility registries
• docs/reconciliation-alpha.md — alpha manifest scope, measurements, and beta handoff
• docs/reconciliation-beta.md — capture and filing resource-fidelity tranche
• docs/reconciliation-complete.md — full mapped-skill review and adoption boundary
• docs/engineering-milestone-alpha.md — integrated acceptance gate and verified result
• docs/ — adaptation decisions, compatibility maps, and runtime audits
• scripts/ — validation, drift, status, and local-brain helpers
• brain/README.md — local brain structure and operating contract
• CONTRIBUTING.md — contribution and validation contract
• SECURITY.md — security reporting and privacy boundary
• CHANGELOG.md — maintainer-facing change history
• docs/release-checklist.md — release gates

Validate the package

python3 scripts/check_engineering_milestone.py
python3 scripts/check_engineering_milestone.py --upstream
python3 scripts/check_engineering_milestone.py --upstream \
  --gstack-repo /path/to/gstack \
  --gbrain-repo /path/to/gbrain

The first command is the local CI gate. The second adds network-backed drift checks for GStack, GBrain, the public Praneet extensions, and Impeccable. The third also rebuilds the complete reconciliation from its pinned Git blobs and is required before a parity claim or release. See the engineering milestone alpha record for the exact acceptance surface and optional GBrain CLI behavior.

Current Codex behavior is checked against the official Codex manual before compatibility or runtime guidance is updated. See AGENTS.md and the documentation refresh log.

Upstream and attribution

• garrytan/gstack
• garrytan/gbrain
• pbakaus/impeccable
• GStack and GBrain adaptations use their MIT-licensed sources. Impeccable-derived guidance retains Apache-2.0 attribution. See LICENSE and NOTICE.

The upstream freshness record separates the last reviewed source commits from the conservative baselines used by the drift checker.

• GStack conservative runtime baseline: 2aa745cb0e4331d683e727ec77385d04cdbb45a2; all mapped skills were reviewed through 94993f74012782fd94416dd44b8314f6363a13a4 on August 8.
• GBrain conservative runtime baseline: b7e3005b5b3f1b54082f9c5990482ebf81a4a807; all mapped skills were reviewed through 0b47afbf402a4e27a648bb9d131ce584461461ea on August 8.
• Impeccable capability baseline: 8259c28209b92792005cec14dad573df39f68eaf; the complete command, detector, hook, live-mode, provider, and behavior-test surface was classified on July 16.

The baseline pins stay conservative so scripts/check_upstream_drift.py continues to reveal broad runtime movement. Per-skill adoption uses source_commit; per-skill freshness uses reviewed_commit or the map's skill_reviewed_commit. This keeps a reviewed deferral distinct from an accepted source update. See the complete reconciliation.