Gentleman-Programming/gentle-ai

Pre-flight Checklist I have searched existing issues and this is not a duplicate I understand that PRs will be rejected if the linked issue does not have status:approved 🔍 Affected Area Documentation 💡 Problem Statement Parent tracker #504 intends to install 11 native GitHub Copilot phase agents under ~/.copilot/agents/, preferably as thin wrappers around the SDD skills already installed under ~/.copilot/skills/. The repository does not yet have organic runtime evidence that one wrapper contract works on both the current stable Windows runtime and a current stable POSIX runtime. Without that evidence, activating production assets would guess at file discovery, frontmatter selectors, skill resolution, delegation, and caller-model inheritance. This issue must establish the wrapper contract before any production capability or asset activation. 📦 Proposed Solution Create a focused, sanitized runtime-proof report for one candidate thin-wrapper format. Test the exact same candid

Pre-flight Checklist I have searched existing issues and this is not a duplicate I understand that PRs will be rejected if the linked issue does not have status:approved 📝 Bug Description In Gentle AI v2.5.0, the canonical sdd-apply skill contains separate model-capable and model-small sections. The model-capable section explicitly includes: Apply any `rules.apply` from `openspec/config.yaml` However, the model-small section contains no equivalent instruction. Consequently, an sdd-apply skill rendered for a small-model profile is not explicitly required to apply the project's phase-specific rules.apply instructions. This is related to, but distinct from, #4114: #4114: sdd-verify omits rules.verify from both model tiers. This issue: sdd-apply includes rules.apply in model-capable but omits it from model-small. 🔄 Steps to Reproduce Check out Gentle AI v2.5.0. Open: internal/assets/skills/sdd-apply/SKILL.md Search for the model-section markers and rules.apply: rg -n '<

Pre-flight Checklist I have searched existing issues and this is not a duplicate I understand that PRs will be rejected if the linked issue does not have status:approved 📝 Bug Description TestInjectClaudeWorkspaceIsDiscoveredByNativeClaudeMCPList (internal/components/mcp/inject_test.go) fails against Claude Code CLI 2.1.114. The test injects a Claude workspace with a user-scope context7 server and an unapproved project-scope server, runs the real claude mcp list inside the workspace, and asserts the output contains both context7 and pending/approval wording for the project server. With CLI 2.1.114 the context7 half passes (reported as connected) but the unapproved project server is never mentioned, so the second assertion fails. This needs triage between two possible roots: (a) the CLI's mcp list reporting contract changed — unapproved project servers are no longer listed with that wording — so the test is version-fragile and should pin to documented CLI behavior or gate on

Pre-flight Checklist I have searched existing issues and this is not a duplicate I understand that PRs will be rejected if the linked issue does not have status:approved 📝 Bug Description The gentle-ai:codegraph-guidance block injected into agent configs instructs agents to create Git worktrees "under the user's home directory, preferably as a sibling such as <repo-parent>/<repo-name>-worktrees/<worktree-name>". For nested project layouts (e.g. ~/Projects/<lang>/<repo>), <repo-parent> is undefined: it can be read as the immediate parent of the repository root (~/Projects/<lang>) or as the user's projects root (~/Projects). The word "preferably" makes the rule advisory, so different agent sessions resolve the placeholder differently. Verified on installed 2.5.0-rc.2 and on current main (internal/components/communitytool/codegraph_guidance.go — identical wording). 🔄 Steps to Reproduce Install gentle-ai with CodeGraph guidance injected into agent configs (OpenCode/Pi). Have a

Pre-flight Checklist I have searched existing issues and this is not a duplicate I understand that PRs will be rejected if the linked issue does not have status:approved 📝 Bug Description skills/gentle-ai-collab-perfect/SKILL.md teaches contributors to recount GitHub pull-request scope with: git diff --stat <base>..<head> That two-dot comparison compares the endpoint trees. When the base advances independently, it includes upstream-only changes as reverse deletions or modifications and therefore inflates or misrepresents the changes introduced by the pull request. GitHub pull-request scope is merge-base based. The contributor guidance should use: git diff --stat <base>...<head> or an explicit merge-base comparison that preserves <head> as the compared side. This is a contributor-facing documentation defect, independent of the implementation of native review authority. 🔄 Steps to Reproduce Check out the current main branch. Inspect skills/gentle-ai-collab-perfect/SKILL.md