Gpd Set Profile

Routes to the set-profile workflow which handles:

• Argument validation (deep-theory/numerical/exploratory/review/paper-writing)
• Config file creation if missing
• Profile update in config.json
• Confirmation with profile description display
• Reminder that $gpd-settings handles concrete runtime model IDs per tier </objective>

<execution_context>

<required_reading> Read all files referenced by the invoking prompt's execution_context before starting. </required_reading>

if $ARGUMENTS.profile not in ["deep-theory", "numerical", "exploratory", "review", "paper-writing"]:
  Error: Invalid profile "$ARGUMENTS.profile"
  Valid profiles: deep-theory, numerical, exploratory, review, paper-writing
  EXIT

/home/qol/.gpd/venv/bin/python -m gpd.runtime_cli --runtime codex --config-dir ./.codex --install-scope local config ensure-section
INIT=$(/home/qol/.gpd/venv/bin/python -m gpd.runtime_cli --runtime codex --config-dir ./.codex --install-scope local init progress --include state,config)
if [ $? -ne 0 ]; then
  echo "ERROR: gpd initialization failed: $INIT"
  # STOP — display the error to the user and do not proceed.
fi

This creates .gpd/config.json with defaults if missing and loads current config. </step>

Update model_profile field:

{
  "model_profile": "$ARGUMENTS.profile"
}

Write updated config back to .gpd/config.json. </step>

Profile set to: $ARGUMENTS.profile

Agents will now operate in this mode:

[Show profile details for selected profile]

Profile definitions:

deep-theory Focus: Rigorous analytical derivations, formal proofs, exact results

Best for: Deriving new results, proving identities, establishing exact relations, formal perturbation theory, renormalization group calculations.

Behavioral highlights: Verification checkpoints after every derivation step. Planner inserts derivation checkpoint every 2 steps. Run the full relevant universal verifier registry plus any required contract-aware checks. All 16 plan dimensions checked. Paper-writer includes proofs in main text (not appendices). Full investigation debugging with formal proof of root cause. Pair with execution.review_cadence=dense when you want more bounded review stops, but set-profile does not change cadence automatically.

numerical Focus: Computational implementation, optimization, convergence, performance

Best for: Implementing solvers, running simulations, optimizing code, convergence studies, parallelization, data pipeline construction.

Behavioral highlights: Convergence testing task added to every numerical computation. Grid/basis/timestep refinement required before results accepted. Richardson extrapolation automatic. Plan-checker emphasizes numerical stability and error budgets. execution.review_cadence stays independent; adaptive is usually the right default unless you explicitly want denser review gates.

exploratory Focus: Rapid prototyping, hypothesis testing, parameter space exploration

Best for: Early-stage investigation, scanning parameter spaces, testing new ideas, order-of-magnitude estimates, dimensional analysis, building intuition.

Behavioral highlights: 3-4 tasks per plan, larger tasks (up to 90 min), only final results verified (skips intermediate checkpoints). Plan-checker reduced to 9 core dimensions. Quick-triage debugging (max 2 rounds). Verifier runs only dimensional analysis + limiting cases + spot-checks + plausibility. Use execution.review_cadence=sparse or adaptive if you want fewer bounded review stops, but required correctness gates still remain.

review (default) Focus: Critical assessment, error checking, literature comparison

Best for: Pre-submission review, debugging wrong results, resolving discrepancies, preparing referee responses, validating collaborator work.

Behavioral highlights: Exhaustive debugging documentation. All verifier checks run plus cross-validation against 2+ literature values. Plan-checker runs all 16 dimensions plus testability checks. Every step cross-references the literature source it implements. This profile often benefits from execution.review_cadence=dense, but cadence remains a separate setting.

paper-writing Focus: Clear exposition, LaTeX production, figure generation, narrative flow

Best for: Writing manuscripts, preparing talks, generating figures, formatting for journal submission, writing supplementary material.

Behavioral highlights: Plans organized by paper sections with tasks mapped to figures, tables, and equations. Narrative-focused execution with clean intermediate expressions. Publication-readiness verification (figures match data, notation consistent, all symbols defined). All 16 plan dimensions checked with emphasis on publication readiness. Full BibTeX formatting against target journal style. Rapid first drafts with multiple revision passes. execution.review_cadence=adaptive or sparse usually fits, but cadence is not profile-owned.

Review cadence interaction:

set-profile changes abstract tier assignments and behavior depth. It does NOT rewrite execution.review_cadence.

• dense: more bounded review stops during execution
• adaptive (default): inject first-result and risky-fanout gates while letting clean segments continue
• sparse: fewest bounded review stops beyond the required correctness gates

Change cadence with $gpd-settings or by editing .gpd/config.json (execution.review_cadence: "dense" / "adaptive" / "sparse").

If you also want to pin concrete runtime model strings for tier-1, tier-2, or tier-3, use $gpd-settings. set-profile changes the abstract tier assignments, not the runtime-native model IDs.

For full agent tier assignments across all 23 agents, see references/orchestration/model-profiles.md. For detailed behavioral effect descriptions per agent per profile, see the "Behavioral Effects" section in references/orchestration/model-profiles.md.

Next spawned agents will use the new profile.

</step>

</process>

<success_criteria>
- [ ] Argument validated against five physics research profiles
- [ ] Config file ensured
- [ ] Config updated with new model_profile
- [ ] Confirmation displayed with profile details including model assignments and behavioral emphasis
</success_criteria>

<!-- [end included] -->

</execution_context>

<process>
**Follow the set-profile workflow** from `@./.codex/get-physics-done/workflows/set-profile.md`.

The workflow handles all logic including:

1. Profile argument validation
2. Config file ensuring
3. Config reading and updating
4. Profile description table generation from RESEARCH_PROFILES
5. Confirmation display showing active profile and its characteristics
   </process>