feat(agents): add named research subagents to Task Researcher#2165

Draft

engineersamuel wants to merge 47 commits into

microsoft:mainfrom

engineersamuel:lucid-veil

Commits on Jun 29, 2026

feat(agents): add task research lanes

authored and
Samuel Mendenhall
committed
feat(agents): make researcher subagent lane aware

authored and
Samuel Mendenhall
committed
docs(prompts): document task research lane mode and subagent inputs

authored and
Samuel Mendenhall
committed
test: add task researcher advisory agent-behavior evals

authored and
Samuel Mendenhall
committed
test(evals): add task researcher comparison fixtures

authored and
Samuel Mendenhall
committed
test(evals): add task researcher static comparison metrics

authored and
Samuel Mendenhall
committed
test(evals): add deepeval task researcher grading

authored and
Samuel Mendenhall
committed
fix(evals): remove egg-info artifacts and improve deepeval_metrics code quality

authored and
Samuel Mendenhall
committed
test(evals): add task researcher comparison reporting

authored and
Samuel Mendenhall
committed
fix(evals): add error handling for TASK_RESEARCHER_RUNNER in capture.py

authored and
Samuel Mendenhall
committed
docs(evals): document task researcher comparison suite

authored and
Samuel Mendenhall
committed
fix(evals): specify pytest test directory to prevent repository-wide collection

authored and
Samuel Mendenhall
committed
fix(evals): align task researcher comparison validation

authored and
Samuel Mendenhall
committed
fix(evals): close task researcher comparison review gaps

authored and
Samuel Mendenhall
committed
feat(scripts): add local Copilot plugin installer

authored and
Samuel Mendenhall
committed
fix(agents): use scalar models for subagents

authored and
Samuel Mendenhall
committed
feat(agents): add named research subagents

authored and
Samuel Mendenhall
committed
feat(agents): use named research lanes

Samuel Mendenhall
authored and
Samuel Mendenhall
committed
feat(collections): add lane research subagents

authored and
Samuel Mendenhall
committed
test(evals): assert named task-researcher fanout

authored and
Samuel Mendenhall
committed
fix: address task-researcher review findings

Samuel Mendenhall
authored and
Samuel Mendenhall
committed
docs(superpowers): add corrected task researcher plan

authored and
Samuel Mendenhall
committed
fix(docs): remove unintended task researcher plan

authored and
Samuel Mendenhall
committed
feat(agents): make named lane subagents read-only synthesis inputs

authored and
Samuel Mendenhall
committed
chore(agents): clarify lane subagent outputs

authored and
Samuel Mendenhall
committed
chore(instructions): restore researcher return symmetry

Samuel Mendenhall
authored and
Samuel Mendenhall
committed
feat(instructions): add untrusted content boundary

authored and
Samuel Mendenhall
committed
feat(evals): correct stimuli, scoring, and fixtures for local lane model

authored and
Samuel Mendenhall
committed
fix(scripts): update task researcher mode compliance

Samuel Mendenhall
authored and
Samuel Mendenhall
committed
fix(scripts): remove shell injection from capture runner

authored and
Samuel Mendenhall
committed
fix(evals): harden task researcher capture handling

authored and
Samuel Mendenhall
committed
feat(scripts): harden local plugin installer with ID validation and path guards

authored and
Samuel Mendenhall
committed
fix(scripts): avoid dry-run directory creation

Samuel Mendenhall
authored and
Samuel Mendenhall
committed
fix(scripts): resolve candidate and install root via pwd -P in safe_installed_path

authored and
Samuel Mendenhall
committed
fix(security): restore model arrays

authored and
Samuel Mendenhall
committed
chore(evals,plugins): regenerate plugin and eval outputs for named subagents

authored and
Samuel Mendenhall
committed
fix(skills): wrap bare URL in fixture output with markdown link

authored and
Samuel Mendenhall
committed
fix(evals): remove inverted noise-control word-match in static_metrics

authored and
Samuel Mendenhall
committed
fix(evals): add fixture frontmatter

authored and
Samuel Mendenhall
committed
fix(agents): add fallback models to lane subagents

authored and
Samuel Mendenhall
committed
fix(prompts): simplify task research controls

authored and
Samuel Mendenhall
committed
fix(agents): restore lane subagent dispatch by dropping model field

authored and
Samuel Mendenhall
committed
refactor(agents): drop redundant task researcher mode-selection block

authored and
Samuel Mendenhall
committed
fix(evals): align task researcher comparison rubric and metric names

authored and
Samuel Mendenhall
committed
style(scripts): reflow plugins readme table

authored and
Samuel Mendenhall
committed
chore(settings): allow deepeval and fanout in spell dictionary

authored and
Samuel Mendenhall
committed
fix(evals): conform advisory tags to string type and regenerate outputs

authored and
Samuel Mendenhall
committed