Declare the skill ID
Add the new ID to the TeamAgentSkillId union in packages/domain/src/team-agent-work.ts. Every other touchpoint keys off this string, so the compiler will then point you at what is missing.
Builder reference
How To Add A Skill.
The places you touch today to add a Jessie skill, in order — and where this process is heading.
Adding A Skill, Step By Step
There is no single registry today. A new Jessie skill is wired in across seven places in three packages and one app. Touch them in this order — TypeScript will flag most of the downstream spots once you do step one.
Register eligibility
In defaultTeamAgentEligibilityConfig (same file): add the Asana tag name, the @mention token(s), and the ID to allowedSkills — the three signals that let a task reach the skill.
Add intake routing and required inputs
Map the skill in skillIdFromMention and the looksLike*Request chain (order-sensitive: specific before broad), and add a required-input block in assessTeamAgentReadiness so Jessie asks for what it needs before running.
Add the runtime policy
Add a RUNTIME_POLICIES entry in packages/domain/src/team-agent-skill-runtime-policy.ts (LLM policy, client-context policy, required LLM stages, initial context tools, insight contract). If it uses the AI Gateway narrative path, also extend the narrative switches in the Worker index.ts.
Wire the executor
Add a dispatch branch in packages/workflows/src/team-agent-execution.ts and, for live data, in executeReadyTeamAgentRun in the Worker. Channel-report skills also get a *-draft.ts and *-html.ts pair modelled on seo-reporting.
Write the SKILL.md contract
Create skills/<skill-id>/SKILL.md with name + description frontmatter and Operating Rules / Inputs / Workflow / Outputs, using the existing channel skills as templates.
Update the tests
Extend team-agent-work.test.ts (eligibility + readiness), team-agent-skill-runtime-policy.test.ts (add to insightSkills), and team-agent-execution.test.ts (dispatch to a review-ready artifact).
Where This Is Heading
Proposed Touching seven places to add one skill is known friction. The agent-loop and typed-tool-library direction (ADR-0019 and Claude_06) proposes collapsing this into a declarative registry and an executable SKILL.md, so adding a skill becomes writing one well-formed descriptor rather than editing seven files. Until that lands, the checklist above is the authoritative guide.