docs: add expected-behavior specs for all 9 factory agents

[gx-ai-architect]

Summary

Closes #788

Creates detailed expected-behavior.md files for all 9 factory agents. These docs serve as diagnostic baselines — when an agent misbehaves, compare its execution trace against the expected behavior doc to pinpoint exactly where it diverged.

Files created (9 docs, ~2,500 lines total)

File	Lines	Workflows covered
docs/expected-behaviors/ceo.md	418	All 8 (Build, Design, Discover, Review, Improve, Research, Refine, Meta)
docs/expected-behaviors/researcher.md	302	Build, Improve, Research, Meta
docs/expected-behaviors/strategist.md	354	Build, Design, Improve, Research, Meta
docs/expected-behaviors/builder.md	315	Build, Design, Improve, Research, Refine, Meta
docs/expected-behaviors/qa.md	301	Improve, Research, Refine
docs/expected-behaviors/archivist.md	382	Build, Design, Improve, Research, Refine, Meta
docs/expected-behaviors/failure-analyst.md	136	Research
docs/expected-behaviors/refiner.md	160	Refine
docs/expected-behaviors/profiler.md	145	Cross-cutting

Each doc contains

1. Identity & Responsibility — what the agent IS vs IS NOT, relationship to other agents
2. Per-Workflow Behavior — for each workflow: phase, inputs, ordered steps, outputs, handoffs
3. Invariants — hard rules (MUST/NEVER/ALWAYS) to check first in any trace
4. Constraints & Forbidden Actions — exhaustive list of what the agent must not do
5. Failure Modes & Diagnostic Signals — table format with trace signals for each known failure
6. Interaction Protocol — output format, CEO review criteria

Verified against issue #783

The CEO doc explicitly covers the build-mode respawn loop bug (#783) with 3 distinct trace signals:

• results.tsv header-only after build phases completed
• Respawn input reporting "0/N phases complete" when git log shows N commits
• current.md overwritten outside Strategist phase

QA verification

All docs verified against actual agent prompts and playbooks:

• 9/9 accuracy (all claims match actual prompt files)
• 8/9 completeness (builder.md Design workflow added in fix commit)
• Cross-agent consistency verified (handoff descriptions match across docs)
• Evaluator/QA naming ambiguity documented with clarification note

🤖 Generated with Claude Code

[github-actions]

Sentrux Quality Report

Absolute

Scanning ....
[scan] git ls-files: 307 total, 295 kept, 12 dropped (ext:12, meta:0, big:0)
[build_project_map] 295 files, 53 unique dirs, 49 cache misses, 2.0ms
[resolve] 434 resolved, 748 unresolved (of 1182 total specs)
[resolve_imports] project_map 2.1ms, suffix_idx 0.6ms, suffix_resolve 7.0ms, total 9.7ms
[build_graphs] 295 files | maps 0.8ms, imports 9.7ms, calls+inherit 2.5ms, total 13.1ms | 433 import, 4356 call, 0 inherit edges
sentrux check — 2 rules checked

Quality: 4706

✗ [Error] max_cc: 5 function(s) exceed max cyclomatic complexity of 30
    factory/cli.py:cmd_ceo (cc=78)
    factory/study.py:study_project_local (cc=43)
    factory/cli.py:_welcome_wizard (cc=39)
    factory/cli.py:cmd_run (cc=36)
    factory/workflow/validation.py:validate_workflow (cc=31)

✗ 1 violation(s) found

Diff (vs base branch)

Scanning ....
[scan] git ls-files: 307 total, 295 kept, 12 dropped (ext:12, meta:0, big:0)
[build_project_map] 295 files, 53 unique dirs, 49 cache misses, 2.1ms
[resolve] 434 resolved, 748 unresolved (of 1182 total specs)
[resolve_imports] project_map 2.2ms, suffix_idx 0.6ms, suffix_resolve 7.1ms, total 9.8ms
[build_graphs] 295 files | maps 0.8ms, imports 9.9ms, calls+inherit 2.5ms, total 13.2ms | 433 import, 4356 call, 0 inherit edges
sentrux gate — structural regression check

Quality:      4706 -> 4706
Coupling:     0.75 → 0.75
Cycles:       4 → 4
God files:    0 → 0

Distance from Main Sequence: 0.35

✓ No degradation detected

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 86.78%. Comparing base (04ce092) to head (bf7a4cf).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #790   +/-   ##
=======================================
  Coverage   86.78%   86.78%           
=======================================
  Files          80       80           
  Lines       12134    12134           
=======================================
  Hits        10531    10531           
  Misses       1603     1603           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.