CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

agent-tuning

Use when changing agent model or effort configuration, per-run Pi profile pins, adapter mappings, or eval candidate profiles.

Quality

60%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./.agents/skills/agent-tuning/SKILL.md
SKILL.md
Quality
Evals
Security

Unified Agent Tuning (internal/agentcfg)

  • agentcfg is the single owner of the harness-neutral model/effort surface and of the mapping down to each harness's native mechanism (claude/copilot --effort, codex -m + -c model_reasoning_effort, grok --reasoning-effort, pi --thinking, opencode's session model/variant, acpx --model for cursor/acp:<target>). OpenCode 2 maps the profile into the session-creation model reference; OpenCode 1 keeps its session-message mapping. Add a harness there, not in an adapter or in eval. rovodev and antigravity are deliberately declared unmappable, so a request for them is a config error rather than a flag that is silently ignored.
  • agent.NewWithOptions is the one funnel: it validates Options.Profile and splices the mapped args after the operator's raw agent_args_override args, so both the pipeline (cfg.AgentProfileFor) and eval replay (Candidate.Profile()) reach every harness by the same path. Never re-derive a model or effort flag at a call site.
  • Precedence is fixed: a raw agent_args_override flag that already pins a knob natively wins and the mapped value is not emitted, which is what keeps every pre-agent_config configuration byte-identical and stops a harness receiving one knob twice. agent_config is global-only for the same reason as agent_args_override.
  • Opt-in Pi run pins (axi run / rerun --model, --effort) resolve in config/pi_profile.go, persist immutably in runs.pi_profile, and are applied before agent construction on launch and recovery. Raw selection flags conflict at creation; later raw/role/global model changes cannot replace a run pin. Unpinned runs retain the precedence above. Semantics owner: docs/src/content/docs/reference/global-config.md (Per-run Pi profiles); regressions: *_pi_profile* / pi_profile_test.go in config, db, daemon, CLI and e2e.
  • Eval candidates are agent,model=<model>[,effort=<level>] (the previous agent+model spelling is refused with a migration message), effort is part of the persisted candidate identity, and agentNeutralGlobalConfig strips agent, agent_args_override, and agent_config so a replay never inherits the capturing machine's pins.
  • Keep eval replay identity comparison centralized in agentcfg.ServedMatchesRequested; do not compare an adapter's reported model directly at call sites. The user-facing normalization semantics and candidate guidance live in docs/src/content/docs/reference/eval.md.
  • Regressions: internal/agentcfg, internal/agent/profile_test.go, internal/config/config_agent_config_test.go, internal/daemon/pipeline_agent_profile_test.go, TestParseCandidate*, TestReplayPinsCandidateModelAndEffortOnTheHarness, TestCaptureStripsEveryHarnessPinFromThePinnedConfig, TestServedMatchesRequested, TestReplayPiModelIdentityComparison.

Per-Run Pi Profile Pin

  • A per-run Pi pin (axi run / rerun --model/--effort) is immutable once the run exists (runs.pi_profile, plus an immutability trigger) and must be fully validated before it supersedes an active validation: startRunWithIntentSourceLocked resolves model/effort against global defaults and agent_args_override.pi, then validates agent eligibility only after the freshly fetched trusted default-branch policy produces the effective repository agent selection (ResolvePiProfile, validatePiProfileAgentsBeforeCancel). Under the default trust policy this lets a trusted project agent: pi replace a global auto or non-Pi scalar/list, while the existing allow_repo_commands opt-in instead delegates repository agent selection to the pushed copy; effective non-Pi fallbacks and review_agents are still refused, except reviewer entries replaced by an explicit per-run reviewer pin. The post-merge ValidatePiProfileAgents/ApplyPiProfile pass at worktree setup and ApplyPiProfile in loadRecoveredConfig remain the enforcement after cancellation. User semantics are owned by docs/src/content/docs/reference/global-config.md (Per-run Pi profiles). Regressions: internal/config/pi_profile_test.go, internal/daemon/pi_profile_test.go (TestPiProfileTrustedRepoPiOverrideAcceptsGlobalAuto, TestPiProfileInvalidLaunchDoesNotSupersedeActiveRun, TestPiProfileTrustedRepoAgentOverrideDoesNotSupersedeActiveRun, TestPiProfileRecoveryUsesPersistedPinAndLegacyUsesLiveConfig), internal/db/pi_profile_test.go, internal/cli/pi_profile_test.go, e2e TestLivePiProfileTrustedPiOverridesGlobalAuto, TestLivePiProfileFeatureBranchPiCannotOverrideGlobalAuto, and TestPiRunProfileSurvivesGlobalChangesAcrossEveryDuty.
Repository

kunchenguid/no-mistakes

Path

.agents/skills/agent-tuning/SKILL.md

Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.