Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Model Profile Resolution · a5c-ai | DecimalAI

a5c-ai/model-profile-resolution

SkillScore64Maintained

Resolve model profile (quality/balanced/budget) at orchestration start and map agents to specific models. Enables cost/quality tradeoffs by selecting appropriate AI models for each agent role.

agent-toolingby a5c-ai

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

AI Engineer pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/a5c-ai-model-profile-resolution/SKILL.md

decimalai skills pull a5c-ai-model-profile-resolution --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/a5c-ai-model-profile-resolution/SKILL.md https://app.decimal.ai/s/a5c-ai-model-profile-resolution/SKILL.md

Raw file: /s/a5c-ai-model-profile-resolution/SKILL.md ↗ · machine-readable summary at /s/a5c-ai-model-profile-resolution.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 55d ago

—+90% tokens0% turns64% of whole cases pass with skill (14/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-07	✗→✓	▲ Improved	129%	0%
case-08	✗→✓	▲ Improved	16%	0%
case-09	✗→✓	▲ Improved	212%	0%
case-10	✗→✓	▲ Improved	75%	0%
case-11	✗→✓	▲ Improved	52%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 7/31/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

a5c-ai / babysitter

library/methodologies/gsd/skills/model-profile-resolution/SKILL.md

View on GitHub ↗

License MITImported 7/31/2026Version v1

Skill contents

SKILL.md

balanced: Mix of high and mid-tier models. Good quality at moderate cost. Default profile.
budget: Cheapest viable models for all agents. Acceptable quality at minimum cost.

The active profile is stored in .planning/config.json under the profile field, or can be overridden with the --profile flag.

Capabilities

1. Read Active Profile

Load the current profile from configuration:

json
// .planning/config.json
{
  "profile": "balanced",
  ...
}

If no profile is set, default to balanced.

2. Profile Definitions

Each profile maps agent roles to specific models:

yaml
quality:
  orchestrator: claude-opus-4-6
  gsd-planner: claude-opus-4-6
  gsd-executor: claude-opus-4-6
  gsd-verifier: claude-opus-4-6
  gsd-plan-checker: claude-opus-4-6
  gsd-phase-researcher: claude-opus-4-6
  gsd-project-researcher: claude-opus-4-6
  gsd-research-synthesizer: claude-opus-4-6
  gsd-roadmapper: claude-opus-4-6
  gsd-codebase-mapper: claude-opus-4-6
  gsd-debugger: claude-opus-4-6
  gsd-integration-checker: claude-opus-4-6

balanced:
  orchestrator: claude-opus-4-6
  gsd-planner: claude-sonnet-4
  gsd-executor: claude-sonnet-4
  gsd-verifier: claude-sonnet-4
  gsd-plan-checker: claude-sonnet-4
  gsd-phase-researcher: claude-sonnet-4
  gsd-project-researcher: claude-sonnet-4
  gsd-research-synthesizer: claude-sonnet-4
  gsd-roadmapper: claude-sonnet-4
  gsd-codebase-mapper: claude-sonnet-4
  gsd-debugger: claude-sonnet-4
  gsd-integration-checker: claude-sonnet-4

budget:
  orchestrator: claude-sonnet-4
  gsd-planner: claude-haiku-4
  gsd-executor: claude-haiku-4
  gsd-verifier: claude-haiku-4
  gsd-plan-checker: claude-haiku-4
  gsd-phase-researcher: claude-sonnet-4
  gsd-project-researcher: claude-haiku-4
  gsd-research-synthesizer: claude-haiku-4
  gsd-roadmapper: claude-haiku-4
  gsd-codebase-mapper: claude-haiku-4
  gsd-debugger: claude-sonnet-4
  gsd-integration-checker: claude-haiku-4

Note: Budget profile uses Sonnet for research and debugging where quality matters most.

3. Per-Agent Model Mapping

Resolve the model for a specific agent:

resolve(agent: "gsd-executor", profile: "balanced")
-> { model: "claude-sonnet-4", profile: "balanced" }

resolve(agent: "gsd-planner", profile: "quality")
-> { model: "claude-opus-4-6", profile: "quality" }

4. Profile Switching

Switch profiles at runtime:

json
// Before
{ "profile": "balanced" }

// After set-profile quality
{ "profile": "quality" }

Profile switches take effect on the next agent spawn. Already-running agents are not affected.

5. Model Capability Validation

Ensure selected model supports required agent features:

Validate gsd-executor needs:
  [PASS] Code generation: all models support
  [PASS] Tool use: all models support
  [PASS] Long context: all models support 200k
  [PASS] Structured output: all models support

Validate gsd-debugger needs:
  [PASS] Code analysis: all models support
  [PASS] Reasoning: opus-4-6 > sonnet-4 > haiku-4
  [WARN] Complex debugging may benefit from quality profile

6. Cost Estimation

Estimate cost per profile for a workflow:

Workflow: execute-phase (Phase 72, 2 plans, 7 tasks)
  Agent spawns: 2 executor + 1 verifier = 3 agents
  Estimated tokens per agent: ~50,000 input + ~20,000 output

  quality (opus-4-6 for all):
    Input:  3 * 50k * $0.015/1k = $2.25
    Output: 3 * 20k * $0.075/1k = $4.50
    Total: ~$6.75

  balanced (sonnet-4 for agents):
    Input:  3 * 50k * $0.003/1k = $0.45
    Output: 3 * 20k * $0.015/1k = $0.90
    Total: ~$1.35

  budget (haiku-4 for agents):
    Input:  3 * 50k * $0.00025/1k = $0.04
    Output: 3 * 20k * $0.00125/1k = $0.08
    Total: ~$0.12

Note: Cost estimates are approximate and based on typical token usage patterns.

Tool Use Instructions

Resolving Profile

Use Read to load .planning/config.json
Extract profile field (default to balanced if not set)
Return profile name

Mapping Agent to Model

Resolve active profile
Look up agent name in profile mapping
Return model identifier
If agent not in mapping, use profile default

Switching Profile

Use Read to load current config
Validate new profile name (quality/balanced/budget)
Update profile field
Write updated config (handled by gsd-tools or direct edit)

Cost Estimation

Accept workflow description (agent types and counts)
Map agents to models via active profile
Estimate token usage per agent
Calculate cost based on model pricing
Return cost breakdown

Process Integration

All agent-spawning processes call this skill at initialization:
new-project.js - Resolve models for 4+ parallel researchers
plan-phase.js - Resolve model for planner, plan-checker
execute-phase.js - Resolve model for executor, verifier
quick.js - Resolve model for planner, executor
debug.js - Resolve model for debugger
map-codebase.js - Resolve model for 4 parallel mappers
iterative-convergence.js - Resolve model per iteration

Output Format

json
{
  "operation": "resolve|switch|validate|estimate",
  "status": "success|error",
  "profile": "balanced",
  "mapping": {
    "agent": "gsd-executor",
    "model": "claude-sonnet-4"
  },
  "costEstimate": {
    "profile": "balanced",
    "agentCount": 3,
    "estimatedCost": "$1.35",
    "breakdown": {
      "input": "$0.45",
      "output": "$0.90"
    }
  }
}

Configuration

| Setting | Default | Description | |---------|---------|-------------| | profile | balanced | Active model profile | | customMappings | {} | Override specific agent-model mappings | | costWarningThreshold | 10.00 | Warn if estimated workflow cost exceeds this |

Error Handling

| Error | Cause | Resolution | |-------|-------|------------| | Unknown profile | Profile name not quality/balanced/budget | Use one of the three valid profiles | | Unknown agent | Agent name not in profile mapping | Use profile default model | | Config not found | .planning/config.json missing | Use default profile (balanced) | | Model deprecated | Mapped model no longer available | Update profile mapping to current model |

Constraints

Only three built-in profiles: quality, balanced, budget
Profile changes take effect on next agent spawn only
Cost estimates are approximate (actual costs depend on token usage)
Custom agent-model overrides in config take precedence over profile defaults
Model availability depends on the runtime environment (Claude Code, OpenCode, etc.)
Budget profile still uses capable models for critical tasks (research, debugging)

Illustrative Want to see what a5c-ai/model-profile-resolution would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

7/31/2026, 23:09 UTC · 22 cases

Measured on 19 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

+59%

64% vs 5%

Agent turns

1.0 vs 1.0

Tokens

+90%

2,703.11 vs 1,423.26

fail → pass · 13pass → pass · 1fail → fail · 8

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I'm setting up a run for the gsd-executor agent. Check `.planning/config.json` to find our current configuration, and figure out which LLM model should handle this role. Please provide the output as a JSON object showing the operation type, status, active profile, and the agent-to-model mapping object.	fail→fail	4,399	4,070	-7%	1	1	0%	230	2,365	+928%	0	0	—
case-02 Please update our project configuration file to use the 'budget' profile for future agent spawns. Output the result in standard JSON format containing the operation name ('switch'), status, and updated active profile.	fail→fail	2,335	4,557	+95%	1	1	0%	350	2,354	+573%	0	0	—
case-03 We are about to execute a phase that spawns 2 executor agents and 1 verifier agent. Based on our active profile settings, please calculate the projected token expenses for this run. Provide the JSON response containing operation ('estimate'), status, profile, total agent count, estimated cost, and the breakdown between input and output charges.	fail→fail	6,382	4,871	-24%	1	1	0%	1,423	2,506	+76%	0	0	—
case-04 When calling the Anthropic Messages API directly in Python using the official anthropic SDK, what model identifier string should be passed in the model parameter to use Claude 3.5 Sonnet?	fail→fail	3,900	2,891	-26%	1	1	0%	784	2,674	+241%	0	0	—
case-05 What is the standard published pricing per million input tokens and output tokens for Anthropic's Claude 3.5 Sonnet model?	fail→fail	3,143	2,746	-13%	1	1	0%	581	2,576	+343%	0	0	—
case-06 In a GSD workspace, under which directory path are individual phase plan and task specification files stored?	fail→fail	8,810	3,470	-61%	1	1	0%	1,517	2,681	+77%	0	0	—
case-07 We need to spawn our primary orchestrator agent under the default balanced profile settings in `.planning/config.json`. Which model should be assigned? Return the result in JSON format showing the operation and mapping.	fail→pass	5,694	2,275	-60%	1	1	0%	1,106	2,530	+129%	0	0	—
case-08 Our repository configuration in `.planning/config.json` has profile set to budget. Determine which LLM model should be assigned to the gsd-debugger agent for an incoming bug investigation. Return the mapping as JSON.	fail→pass	13,000	3,035	-77%	1	1	0%	2,431	2,813	+16%	0	0	—
case-09 The project configuration is set to the budget profile. Resolve the LLM model for the gsd-phase-researcher agent role and return the result in JSON format.	fail→pass	4,693	2,771	-41%	1	1	0%	867	2,709	+212%	0	0	—
case-10 Our configuration `.planning/config.json` uses the budget profile. What model is assigned when spawning a gsd-executor agent? Provide JSON output.	fail→pass	6,576	1,929	-71%	1	1	0%	1,416	2,473	+75%	0	0	—
case-11 If `.planning/config.json` has the profile key configured as quality, which LLM model is selected for the gsd-planner agent? Return JSON output.	fail→pass	8,010	2,062	-74%	1	1	0%	1,645	2,503	+52%	0	0	—
case-12 The file `.planning/config.json` exists in our workspace but does not contain a profile key. Which profile should be used by default when resolving models for agent tasks?	pass→pass	6,016	2,181	-64%	1	1	0%	1,022	2,443	+139%	0	0	—
case-13 A developer attempts to switch our workspace profile to ultra-fast using a profile update request. How should this request be handled? Return a JSON response with operation switch.	fail→pass	3,204	2,729	-15%	1	1	0%	623	2,617	+320%	0	0	—
case-14 We just updated `.planning/config.json` from balanced to quality while two gsd-executor agents are currently executing tasks. What happens to the models running those active agents?	fail→fail	5,610	3,102	-45%	1	1	0%	994	2,643	+166%	0	0	—
case-15 A process requests model resolution for an unlisted custom agent named gsd-custom-worker under the balanced profile. Which model should be selected?	fail→pass	7,885	3,214	-59%	1	1	0%	1,430	2,755	+93%	0	0	—
case-16 Our `.planning/config.json` contains `profile: balanced` and `customMappings: {"gsd-executor": "claude-opus-4-6"}`. Which model is resolved for gsd-executor?	fail→fail	3,456	2,550	-26%	1	1	0%	541	2,523	+366%	0	0	—
case-17 Calculate projected token costs for a workflow that spawns 2 gsd-executor agents and 1 gsd-verifier agent under the budget profile. Provide JSON output formatted with costEstimate.	fail→pass	8,945	5,840	-35%	1	1	0%	2,088	3,443	+65%	0	0	—
case-18 Calculate projected token costs for a workflow spawning 2 gsd-executor agents and 1 gsd-verifier agent under the quality profile. Return JSON with the costEstimate breakdown.	fail→pass	8,040	3,848	-52%	1	1	0%	2,053	2,943	+43%	0	0	—
case-19 Perform a model capability validation check for the gsd-debugger agent when running on the balanced profile. Provide JSON or structured output of the validation result.	fail→pass	10,080	5,418	-46%	1	1	0%	2,011	3,283	+63%	0	0	—
case-20 When attempting to read active profile configuration, `.planning/config.json` is missing from the file system. What profile is adopted?	fail→pass	5,743	1,937	-66%	1	1	0%	768	2,429	+216%	0	0	—
case-21 Resolve the LLM model for the gsd-codebase-mapper agent under the active balanced profile in JSON format.	fail→pass	5,963	2,711	-55%	1	1	0%	979	2,700	+176%	0	0	—
case-22 Under the budget profile in `.planning/config.json`, which model is assigned to the gsd-roadmapper agent? Return JSON output.	fail→pass	17,319	2,367	-86%	1	1	0%	4,186	2,621	-37%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +59 percentage points is the difference between those two pass rates over the 19 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 4.6%
Pass rate with the skill: 63.6%
Cases attempted: 22
Cases counted toward lift: 19
Cases that regressed: 3
Skill version measured: v1
Run completed: 7/31/2026, 23:09 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-07

Prompt

We need to spawn our primary orchestrator agent under the default balanced profile settings in `.planning/config.json`. Which model should be assigned? Return the result in JSON format showing the operation and mapping.

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run64

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+139% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 54 days ago

Other measured skills in the registry, with their headline benchmark lift.