Skills overview

SaaSFoundryAI installs project-owned sf-* skills that explain how to work safely in the generated architecture. They are executable documentation for coding agents: workflow guards, integration grammar, Git operations, SRS procedures and external-tool adapters live beside the code they govern.

Why every project skill starts with sf-

• sf-workflow, not a generic workflow
• sf-git-commit, not a generic commit
• sf-tool-github-projects, not a generic GitHub helper

The prefix prevents a globally installed skill from silently replacing project policy. In a SaaSFoundryAI project, agents prefer the repository's sf-* procedure whenever both a generic and project-specific capability exist.

Skill categories

Category	Installed when	Examples
Core	Every harness installation	sf-git-commit, sf-git-create-pr, sf-utils-fix-errors
Workflow	A workflow is configured	sf-workflow plus the selected board skill
Integration grammar	A generated app is managed	sf-integration-rules
SRS	SRS is enabled	sf-srs and its backend adapter
Optional tool	Selected during setup/update	sf-tool-context7, sf-tool-atlassian, sf-tool-notion, sf-tool-figma

One source, several agent entrypoints

Claude Code remains the compatibility source during the multi-agent transition:

.claude/skills/          # managed source used by Claude Code and guarded scripts
.agents/skills/          # shared copies for Codex and other declared agent profiles
CLAUDE.md                # Claude entrypoint
AGENTS.md                # portable/Codex entrypoint; points back to project rules
GEMINI.md                # Gemini entrypoint when selected

The exact files depend on the declared profiles in .saasfoundry.json. sf agents list, sf agents doctor and sf status --agent-friendly --no-network report declarations and file presence. They do not claim that a host discovered skills, executed hooks or received credentials; verify those behaviors in the actual host.

Generated multirepos place the managed harness where each repository needs it. Monorepos can centralize the shared instruction surface at the root. See Agent coexistence and Project structure for the exact topology.

How an agent uses a skill

1. Read the project's instruction entrypoint.
2. Inspect .saasfoundry.json for topology, modules, agent profiles and workflow.
3. Select the matching sf-* skill from the declared skill root.
4. Read its complete SKILL.md and referenced status/reference files.
5. Execute guarded scripts rather than recreating their side effects manually.

Some hosts support keyword-based discovery or slash commands; others require the agent to read the skill explicitly. SaaSFoundryAI documents both the desired capability and the host limitation instead of pretending every runtime behaves like Claude Code.

Workflow skill and board adapters

sf-workflow owns the delivery semantics. A board skill performs the actual ticket operations.

Adapter	v1 status
sf-tool-github-projects	Complete v1 delivery contract
Jira adapter	Experimental; no promise of full guard or hierarchy parity
Linear adapter	Experimental; no promise of full guard or hierarchy parity
Notion	Complete v1 SRS backend; not a complete workflow tracker

The workflow skill reads the configured status sequence, including the seven-status team preset, the five-status Solo preset or a custom template. It never needs a separate “advanced” workflow skill: complexity controls analysis and review depth inside the one workflow.

Discovery and diagnostics

sf skill list
sf skill describe sf-workflow
sf agents list --json
sf agents doctor codex claude-code
sf status --agent-friendly --no-network

To inspect the managed source directly:

cat .claude/skills/sf-workflow/SKILL.md
cat .agents/skills/sf-workflow/SKILL.md

Use whichever path exists for the declared current host. Do not copy personal/global skills into the repository or infer consent from a discovered directory.

Updates and local customisation

sf update compares managed skill files with the baseline recorded in .saasfoundry.json:

• unchanged managed files can be updated automatically;
• local edits become explicit conflicts instead of being overwritten silently;
• credentials remain outside the repository;
• shared agent declarations and instruction files remain reviewable changes.

If a convention is specific to your product, create an sf-* custom skill in the project and document its ownership. See Creating skills.

Continue

• Core skills
• Tool skills
• Creating skills
• Skills system architecture
• Agent coexistence