Add continuous cloud-agent workflow and repository safeguards

[naytewilson]

What changed

• Added fstack-run, a continuous source-truth-to-verified-PR workflow for cloud agents.
• Preserved the original interactive skills and updated /fstack routing to distinguish interactive and continuous modes.
• Added repository-wide instructions for AGENTS-compatible clients, Claude Code, and GitHub Copilot coding agent.
• Added dependency-free skill validation, GitHub Actions CI, an evidence-focused PR template, and an A-Z cloud-agent operator guide.
• Documented both the official GitHub CLI gh skill route and the broader cross-agent npx skills route.

Why this is the smallest complete change

The existing skills remain intact. One new skill owns continuous execution rather than weakening every interactive skill or adding a large orchestration framework. Validation uses POSIX shell already available on GitHub-hosted runners and introduces no application runtime or package manifest.

Evidence

Branch: feat/cloud-agent-runner
Head: 52d30a1489c5e2ff1334058544ddf24809bf0a32
Base: main@48598b558b0d33de0e4a79afb4824c623ddbeeee
Comparison: 13 commits ahead, 0 behind
Changed files: 10, all task-owned

GitHub Actions run: 30628073142
Conclusion: success

sh -n scripts/validate.sh
Observed: success

sh scripts/validate.sh
Observed: Validated 14 skills.

git diff --check "origin/main...HEAD"
Observed: success, no output

Skill and documentation checks

• sh -n scripts/validate.sh
• sh scripts/validate.sh
• Base-branch whitespace diff check
• New and changed skills documented in README.md
• Frontmatter names match skill directories
• No dependencies, secrets, credentials, private local paths, or generated artifacts added
• Final diff reviewed; inaccurate dependency wording was corrected before the final validation run

Missing evidence or remaining risk

• gh skill publish --dry-run is documented but was not executed by this dependency-free CI workflow.
• Host-level branch rules, secret scanning, code scanning, merge policy, and automatic branch deletion were not changed or independently verified by this PR.
• The PR intentionally does not merge or publish a release.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 52d30a1489

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Compare main pushes against the pre-push commit

On a push to main, GITHUB_BASE_REF is unset, so base becomes main; after checkout, origin/main and HEAD both reference the newly pushed commit. Consequently, git diff --check "origin/main...HEAD" examines an empty range and direct pushes containing whitespace errors pass this workflow. Use the push event's before SHA for push events while retaining the base-branch comparison for pull requests.

AGENTS.md reference: AGENTS.md:L41-L47

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Preserve skill paths when iterating validator results

When the repository is checked out beneath a path containing spaces, unquoted expansion in for file in $files splits every discovered SKILL.md path into multiple words; the first sed invocation then fails on a truncated path and validation exits. Iterate over newline-delimited input without shell word splitting so the advertised portable validator works from ordinary directories such as My Projects/fstack.

AGENTS.md reference: AGENTS.md:L5-L5

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Validate README skill entries in both directions

This check only ensures that every skill still on disk is mentioned somewhere in the README. If a skill directory is removed or renamed while its old README table and count remain, the loop never checks that stale name and the validator reports success, allowing the documented collection and routing information to diverge from what users can install. Compare the README skill set with the on-disk set in both directions.

AGENTS.md reference: AGENTS.md:L29-L29

Useful? React with 👍 / 👎.