Problem It Solves
Closes gap: #173
Memory is never automatically populated. teamwork complete marks a workflow done and writes metrics, but extracts nothing from the completed workflow for future use. Patterns, decisions, and anti-patterns are lost unless a human manually runs teamwork memory add.
Proposed Behavior
teamwork complete <workflow-id> gains an optional (default: on) post-completion analysis step:
$ teamwork complete wf-042
✓ Workflow wf-042 marked complete.
Analyzing completed workflow for memory candidates...
Proposed memory additions (review and accept/edit/reject each):
[1/3] PATTERN — domain: auth
"RS256 JWT with sub-claims for roles: validate aud and iss on every request"
Source: Step 2 (Architect) handoff, confirmed in Step 6 (Reviewer) handoff
> Accept (a), Edit (e), Reject (r), Skip all (s): a
[2/3] ANTIPATTERN — domain: logging
"Never log Authorization headers or request bodies containing tokens"
Source: Step 3 (Security Auditor) finding SA-002
> Accept (a), Edit (e), Reject (r), Skip all (s): e
Edit: "Never log Authorization headers or request bodies — they may contain tokens or PII"
> a
[3/3] DECISION — domain: api
"Auth errors return 401 with WWW-Authenticate header per RFC 7235"
Source: Step 2 (Architect) handoff, ADR-012
> Accept (a), Edit (e), Reject (r), Skip all (s): r
2 entries added to .teamwork/memory/ and committed.
The extraction reads all handoffs for the workflow, identifies: architectural decisions (from Architect/Planner handoffs), security findings (from Security Auditor handoffs), reviewer-flagged patterns (from Reviewer handoffs), and resolved blockers (from state file block records). It proposes them as typed memory entries. Human reviews and approves before anything is written.
--no-memory flag skips the proposal for scripted/CI usage.
Dependencies
Complexity Estimate
Large — 6 files: handoff section extractor, memory proposal engine, interactive TUI prompt (extends internal/tui/tui.go), teamwork complete command integration, duplicate detection, tests
Problem It Solves
Closes gap: #173
Memory is never automatically populated.
teamwork completemarks a workflow done and writes metrics, but extracts nothing from the completed workflow for future use. Patterns, decisions, and anti-patterns are lost unless a human manually runsteamwork memory add.Proposed Behavior
teamwork complete <workflow-id>gains an optional (default: on) post-completion analysis step:The extraction reads all handoffs for the workflow, identifies: architectural decisions (from Architect/Planner handoffs), security findings (from Security Auditor handoffs), reviewer-flagged patterns (from Reviewer handoffs), and resolved blockers (from state file block records). It proposes them as typed memory entries. Human reviews and approves before anything is written.
--no-memoryflag skips the proposal for scripted/CI usage.Dependencies
internal/handoff/handoff.go) extended to support structured section extractioninternal/memory/memory.go)Add()called per approved entryComplexity Estimate
Large — 6 files: handoff section extractor, memory proposal engine, interactive TUI prompt (extends
internal/tui/tui.go),teamwork completecommand integration, duplicate detection, tests