Proposes concrete actions from guarded facts and waits for explicit human approval before any side effect.
Designs non-overlapping agent roles with goals, backstories, and stop conditions for a crew.
Implements on a feature branch only after plan and preview approvals are on record.
Appends who did what: human vs agent, with timestamps and references for later review.
Reads untrusted channels, wraps content as data, and never takes external actions itself.
Builds a safe static/example preview of a feature before touching the main repo.
Gathers structured company signals: product, pricing, hiring, and changelog crumbs.
Turns a career plan into a concrete learning backlog with practice projects.
Designs a reliable schedule of agent jobs with owners, inputs, and failure alerts.
Cross-checks company intel claims against sources and marks contradictions.
Gathers accurate beginner-friendly facts and examples for a tutorial topic with sources.
Reviews a proposed multi-agent graph for loops, token blowups, and missing stop conditions.
Drafts community or social replies for scheduled engagement windows, waiting for approval to post.
Writes explicit task contracts: description, expected output, and handoff schema between roles.
Looks across skills, roles, and compensation to recommend a coherent career positioning plan.
Reviews tutorials for accuracy, missing steps, and beginner confusion; returns a punch list.
Writes detailed build plans and holds a hard gate: no production writes until the user approves.
Scouts roles, companies, and networking angles that match the strategist's positioning.
Writes a structured beginner tutorial from a verified concept fact sheet.
Runs scheduled digests of metrics, inbox, and blockers into a short morning brief.
Assembles verified company sheets into clean one-pager profiles for sales or research use.
Builds the strongest affirmative case for a decision with assumptions made explicit.
Checks every claim against sources in a tight loop until errors are zero or flagged.
Builds a token-budgeted prompt pack for each DAG task with only the files and upstream summaries it needs.
