| name | technical-analysis |
| description | Produce a structured technical analysis with method-level change list, notes, and conservative time estimates. Also find architectural deepening opportunities when user asks for architecture review or refactoring candidates.
|
Technical Analysis Skill
Analyze the user's feature request and produce a structured technical analysis document with method-level details, notes, and estimation.
When to use
/technical-analysis- User says "technical analysis", "estimation", or "breakdown".
- User provides a task link (Phabricator, Jira, GitHub issue) and asks for planning or analysis.
- User asks "how long would this take", "what would need to change", or "where do I start".
- User asks for an architecture review or refactoring candidates ("find deepening opportunities").
Format
Produce output in this exact structure:
### Technical Analysis:
1. <Subtask title and changes grouped by logical change>:
- `<method_name>` at `<relative_file_path>`
- `<method_name>` at `<relative_file_path>`
2. <Subtask title and changes grouped by logical change>:
- `<method_name>` at `<relative_file_path>`
...
### Notes:
1. <Note>
2. <Note>
### Estimation:
[ ] <Subtask title matching numbered item>: <workdays> day(s)
[ ] <Subtask title matching numbered item>: <workdays> day(s)
...
[ ] Full integration testing: <workdays> days
[ ] Buffer for unknowns: <workdays> days
Implementation: <sum> workdays -> ~<weeks> weeks
Testing & buffer: <sum> workdays -> ~<weeks> weeks
Total: <sum> workdays -> ~<weeks> weeks -> ~<sprints> sprints
Rules
- Technical Analysis section: One numbered item per logical change group. Each item has sub-bullets with the exact method name in backticks and the relative file path.
- Notes section: capture constraints, dependencies, side effects, and decisions that affect implementation.
- Estimation section: Use
[ ] (unchecked) so they can be marked done later. One bullet per numbered item above. Add test coverage into each item's estimate. Be conservative, better to finish early than late. Add separate bullets for "Full integration testing" and "Buffer for unknowns". Then split total into "Implementation" and "Testing & buffer" lines, then a "Total" line with workdays, weeks, and sprints.
- Estimations must be safe: A single software engineer working without AI assistance should be able to complete the work within the estimated time. Do not assume AI tooling will speed anything up.
Process
- If given a link, fetch the resource first to extract requirements.
- Ask clarifying questions to understand scope, boundaries, and any hard constraints.
- Search the codebase for related methods, patterns, and files, trace the full flow end-to-end.
- For each identified change point, note the exact method name and file path.
- Group changes into logical numbered items.
- Estimate each item in workdays (be conservative), adding test coverage time.
- Present the result and iterate with the user until approved.
Architecture Improvement
When the user asks for architecture review, refactoring candidates, or "find deepening opportunities", supplement the technical analysis with an architecture review. Use these concepts:
Core concepts
- Module: Anything with an interface and an implementation (function, class, package, slice).
- Interface: Everything a caller must know to use the module (types, invariants, error modes, ordering).
- Deep module: A lot of behavior behind a small interface. High leverage.
- Shallow module: Interface nearly as complex as the implementation. Low leverage.
- Seam: Where an interface lives, a place behavior can be altered without editing in place.
Deletion test
For any module you suspect is shallow, imagine deleting it. If complexity vanishes (was a pass-through), it was not earning its keep. If complexity reappears across N callers (they each reimplement the same logic), it was earning its keep.
Process
- Walk the codebase and note friction: Where does understanding one concept require bouncing between many small modules? Where are modules shallow? Where has code been extracted just for testability but the real bugs hide in how it is called?
- For each friction point, present: Files involved, the problem, a proposed deepening, and the benefits (locality, leverage, testability).
- Rank candidates by impact (Strong, Worth exploring, Speculative).
- Let the user pick which to explore, then drill into the design tree with them.
