On-demandOn-demand: invoked explicitly — never occupies your agent's context until called.
◆Doc Ops pack
Last verified 8/20/2026
What it changes
Passed 9 of 18 test cases with the skill, and 7 of 18 without it.
Without skill38%
With skill50%
Checks it passed that the model alone missed
Includes a confidence level for each assumption entry
Includes a forward reference to the target file to be updated when resolved
Focuses CONTEXT.md on business domain context from a stakeholder perspective
and 2 more in the full report
4 fixed8 passed either way11 still failing1 broke
Pass-rate lift
+13 pp
Tokens
+15%
Agent turns
0%
Response time
-56%
Model gemini-3.6-flashMeasured 8/20/2026v112 of 24 cases passed with skill · 18 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 4 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 4
Prompt
We are populating TODO.md with task items discovered during workspace analysis. What structure should be used for the item headers in TODO.md?
What was checked, without and with the skill
Checked
Without skill
With skill
Includes priority, timing, description, and location in each header using the format priority when what where
Missed
Passed
case-08 · gemini-3.6-flash · 8/20/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md131 linessingle file
<init_workspace_documentation>
<when_to_use_skill>
Workspaces lack structured documentation, forcing every session to re-discover facts and repeat mistakes. This skill creates five foundational docs from source code analysis. Proof: all five docs exist, are non-empty, complementary, and track unknowns.
</when_to_use_skill>
<core_concepts>
All Rosetta prep steps MUST be FULLY completed, load-context skill loaded and fully executed
ACQUIRE reverse-engineering/SKILL.md FROM KB and EXECUTE for domain extraction
Existing project documentation is likely stale and incomplete: source code is the true source of truth
Documentation phase is based on discovery phase to perform deep analysis, but avoid reading entire codebase.
Select which files to read, group organize by modules/batches/groups and must assign to subagents to execute.
</core_concepts>
Dual-mode based on state.mode:
Scan for each target doc file
Compare existing content against codebase findings
install = create all; upgrade = update gaps only
Never overwrite human-added content; merge alongside
Report created/updated/skipped files
Analyze project structure and key source files
Create TODO task per document with business context angle
Track unknowns in ASSUMPTIONS.md with forward references
Create or update documents:
CONTEXT.md:
What this doc is for and what it should contain, self-defining style
Self-defines purpose, content type, style
Bulleted business context, purpose, domain — stakeholder perspective