High-cost orchestrator workflow for large, high-risk, multi-phase coding efforts with meaningful dependencies and review gates. Do not activate for routine multi-file changes.
64
76%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
View guide
Passed
No findings from the security scan
Fix and improve this skill with Tessl
tessl review fix ./src/skills/deepwork/SKILL.mdDeepwork is an orchestrator workflow for heavy coding sessions. Use it only when the work is clearly large or high-risk: multiple dependent phases, cross-cutting architectural change, unsafe-to-partially-ship migration, or sustained coordination across several specialist lanes.
Do not infer Deepwork merely because a task touches multiple files. Do not use it for trivial edits, quick docs changes, simple bug fixes, or routine bounded features.
When deepwork is active, the orchestrator must manage the work as a scheduler, not as the default implementation worker.
Required behavior:
.gitignore and .ignore; add only missing entries, without
duplicates, so .gitignore contains .slim/deepwork/ and .ignore contains
!.slim/deepwork/ and !.slim/deepwork/**;.slim/deepwork/;src/, docs/); reserve
.slim/deepwork/ strictly for progress files;@oracle to review the phase result before continuing;@designer, preserve designer intent across later
phases. Use @fixer only for mechanical follow-up that does not alter the
UI/UX;Oracle reviews are automatic gates between the planned implementation phases. Before dispatch, decide the phases from the task itself: its dependencies, integration boundaries, and meaningful delivery points. Record the phase order, the total review count, the review after each phase, and a short reason for each gate in the deepwork file and compact user overview.
Avoid micro-phases created only to make reviews smaller or cheaper. Larger, complex tasks can have broader phases, broader patches, and correspondingly broader phase reviews. The goal is a sensible number of predictable review gates, not the smallest possible review scope. Never add an extra Oracle review merely to re-confirm a mechanical fixer change.
When a deepwork phase includes @designer, treat the delivered UI/UX as
accepted design intent for later phases. Record any important design decisions in
the deepwork file before continuing.
After designer work:
@designer;@fixer only for bounded mechanical follow-up that preserves the design
exactly, such as wiring, tests, type fixes, or non-visual behavior changes;Create a task-specific file such as:
.slim/deepwork/<short-task-slug>.mdBefore creating this file—and before planning or delegation—inspect the existing
.gitignore and .ignore. Add only missing entries and do not add duplicates:
# .gitignore
.slim/deepwork/# .ignore
!.slim/deepwork/
!.slim/deepwork/**These rules keep deepwork state git-local while allowing OpenCode to read it.
Do not follow a rigid template. Choose whatever markdown structure best fits the work. The file only needs to remain useful as persistent session state and should capture, as applicable:
@librarian to avoid oracle doing its own
research;Update this file after major decisions, valuable specialist research, reviews,
phase completions, validation results, and scope changes.
When @librarian docs, code reads, or external references produce useful
information, reconcile the result and record the accepted findings here so later
planning and reviews share the same context instead of rediscovering it.
Don't put actual contents of local files, reference them by path only.
Use the scheduler model throughout:
1c0e1f4
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.