You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Today the acq/sbx kits clone the playbook repo's .agents/skills into sandboxes (via playbook-clone.sh). The patterns repo's own skills (secure-code-review, the frontend/USWDS set, communications, the pattern-router skill, etc.) and scripts/route_patterns.py are a contributor-local toolchain — they never reach a sandbox/model through the kits approach. The user's goal is to inject/add all our skills easily and surface them to all models via kits.
Proposal (consensus 7/7 approved)
Add a neutral hybrid/v1 acq kit (e.g. acq-kits/agentic-coding-patterns) that, at sandbox startup:
Fetches the patterns repo at a pinned full-SHA ref via the GitHub REST tarball (same credential-substitution-safe path as the playbook kit — NOT git clone), verifying a content sha256 over the skills tree.
Optionally installs scripts/route_patterns.py + INDEX.yaml + schemas/taxonomy.yaml so the deterministic pattern-router skill works in-sandbox.
Records kit-bundle provenance (bundle name, repo, applied ref, timestamp) — reuse the provenance schema added in patterns#273 / quickstart#236 so acq kit check|update can tell a stale patterns-skills bundle from a current one.
Constraints
Router stays an OPTIONAL deterministic selection aid layered on top of standard progressive-disclosure discovery — model access to skills MUST NOT depend on the router (consensus condition).
Pin a full 40-char SHA; never fetch mutable main.
Ensure the linker refuses to overwrite existing skill dirs and does not follow symlinks outside the kit root (security condition).
Guard against name collisions between the playbook's skills and the patterns' skills in the same shared root.
Acceptance
A sandbox provisioned with the kit surfaces the patterns skills to OpenCode/Claude/Codex via their skills roots.
acq kit check reports the patterns-skills bundle status.
Provenance recorded; SHA-pinned; no mutable fetch.
Depends on #277 (name/description conformance) and #278 (flat layout). Relates to quickstart#235/#236 (provenance).
AI-assisted (OpenCode); consensus-reviewed. Requires human review.
Problem
Today the acq/sbx kits clone the playbook repo's
.agents/skillsinto sandboxes (viaplaybook-clone.sh). The patterns repo's own skills (secure-code-review, the frontend/USWDS set, communications, thepattern-routerskill, etc.) andscripts/route_patterns.pyare a contributor-local toolchain — they never reach a sandbox/model through the kits approach. The user's goal is to inject/add all our skills easily and surface them to all models via kits.Proposal (consensus 7/7 approved)
Add a neutral
hybrid/v1acq kit (e.g.acq-kits/agentic-coding-patterns) that, at sandbox startup:git clone), verifying a content sha256 over the skills tree.~/.agents/skills,~/.claude/skills,~/.cursor/skills, per-agent roots) — flat per Skills: nested skill dirs (frontend/, communications/, outreach/, meta/) may not surface under flat ~/.agents/skills discovery #278.scripts/route_patterns.py+INDEX.yaml+schemas/taxonomy.yamlso the deterministicpattern-routerskill works in-sandbox.acq kit check|updatecan tell a stale patterns-skills bundle from a current one.Constraints
main.Acceptance
acq kit checkreports the patterns-skills bundle status.Depends on #277 (name/description conformance) and #278 (flat layout). Relates to quickstart#235/#236 (provenance).
AI-assisted (OpenCode); consensus-reviewed. Requires human review.