[PM-33571] llm: Add requirements refinement and planning skills#2445
[PM-33571] llm: Add requirements refinement and planning skills#2445SaintPatrck merged 2 commits intomainfrom
Conversation
SaintPatrck
force-pushed
the
llm/planning-skills
branch
from
9e6d14b to
a0c5553
Compare
github-actions
bot
added
the
t:llm
|
Claude finished @SaintPatrck's task in 4m 4s —— View job I'll analyze this and get back to you. |
|
Great job! No new security vulnerabilities introduced in this pull request |
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #2445 +/- ##
==========================================
- Coverage 87.09% 85.95% -1.14%
==========================================
Files 1859 2089 +230
Lines 164348 179153 +14805
==========================================
+ Hits 143132 153995 +10863
- Misses 21216 25158 +3942 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
fedemkr
left a comment
fedemkr
left a comment
There was a problem hiding this comment.
Looks good, just some minor tweaks.
| ## Phase 3: Plan Implementation | ||
|
|
||
| Once requirements are confirmed, invoke the `planning-ios-implementation` skill to: | ||
| - Classify the change type |
There was a problem hiding this comment.
❓ Maybe it's defined in another place but which would be the possible change types? Or is it just general for the LLM to figure out?
There was a problem hiding this comment.
Classifying is a step in the planning-ios-implementation skill. This is "priming" Claude so it knows what the skill will be doing. I see that planning-ios-implementation doesn't mirror the Android version, though, so I'll fix that up.
| ## Prerequisites | ||
|
|
||
| - Requirements must be clear. If not, invoke `refining-ios-requirements` first. | ||
| - Read `Docs/Architecture.md` before proceeding — it is the authoritative source for all patterns. | ||
|
|
There was a problem hiding this comment.
🤔 Should it also read the Testing.md file as prerequisite? As sometimes having the structure to be testable changes how it's implemented.
There was a problem hiding this comment.
I've not run into that as an issue on the Android side of things. I would wait until we see Claude exhibiting undesirable behavior before preemptively telling it to consume more context. If that turns out to be needed, I would prefer a simple directive like, "consider testability while designing implementation." before having it re-ingest the entire Testing.md.
- refining-ios-requirements skill: ingests Jira/Confluence via MCP, extracts requirements, performs gap analysis (error states, edge cases, extensions, accessibility), maps to iOS domain (Auth/Autofill/Platform/Tools/Vault) - planning-ios-implementation skill: classifies change type, explores existing patterns, lists files to create/modify with domain placement, produces dependency-ordered implementation phases, assesses risk - /plan-ios-work command: orchestrates both skills end-to-end with user confirmation between phases, saves design doc to .claude/outputs/plans/ - CLAUDE.md: compress directory tree listing, remove Auth/Authorization detail section (now in refining-ios-requirements skill), remove redundant Core Directives list, replace external style URL with local Docs/ references
…e extension and TOTP detail
SaintPatrck
force-pushed
the
llm/planning-skills
branch
from
a0c5553 to
ec645fc
Compare
Bitwarden Claude Code ReviewOverall Assessment: APPROVE This PR adds two Claude skills ( No findings were identified. The changes are documentation and LLM configuration only, with no impact on application code, security, or build behavior. |
|
@fedemkr I replaced the various domain lists with a single reference in |
3 participants
🎟️ Tracking
https://bitwarden.atlassian.net/browse/PM-33571
📔 Objective
Adds two new Claude skills and a command for the planning phase of iOS feature development.
What changed:
skills/refining-ios-requirements— Ingests Jira/Confluence tickets via MCP, extracts structured requirements and acceptance criteria, performs gap analysis (error states, edge cases, multi-account, extension memory, accessibility), and maps work to the correct iOS domain (Auth/Autofill/Platform/Tools/Vault). Produces a structured requirements doc and pauses for user confirmation before proceeding.skills/planning-ios-implementation— Classifies the change type (new feature, enhancement, service, bug fix), explores existing codebase patterns, lists all files to create/modify with domain placement, produces dependency-ordered implementation phases (Core → Services → Processors → Views → DI → Tests), and assesses risks (security, extensions, multi-account, SDK)./plan-ios-workcommand — Orchestrates both skills end-to-end: ingests a ticket via MCP, refines requirements, confirms with user, plans implementation, and saves the design doc to.claude/outputs/plans/<ticket-id>.md.CLAUDE.md — Compresses the directory tree listing from 45 to 5 lines; removes the Auth/Authorization detail section (now discoverable via the
refining-ios-requirementsskill); removes the redundant Core Directives numbered list (covered by the DO/DON'T section andDocs/pointers); replaces the external style guide URL with localDocs/Architecture.mdandDocs/Testing.mdreferences.