lfd-design

Design a loss function and harness for a long-running /goal optimization run (loss-function development, LFD). Use when the user wants to set up an autonomous optimization loop, distill a product from public artifacts, turn a spec into an optimization target, or asks to design a /goal. Observes the existing environment, interrogates the task, ingests or generates the spec, builds a blinded eval, generates and verifies the harness, red-teams the target for cheats, and emits goal.md ready to launch. Re-invoke in patch mode when a running loop cheated and the loss function needs patching.

Quality

—

Does it follow best practices?

Impact

—

No eval scenarios have been run

Securityby

Advisory

Suggest reviewing before use

Quality

Content

100%

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The content is a tightly written, fully sequenced design playbook with concrete artifacts, validation feedback loops, and one-level-deep references to real bundle files. It earns the top anchor on each dimension and does not fall to 2 because the guidance is concrete and the references are real rather than implicit.

Dimension	Reasoning	Score
Conciseness	The body is dense and assumes Claude's competence, with no padding about what an eval or optimizer is; every line carries design doctrine, fitting the lean-and-efficient anchor.	3 / 3
Actionability	Concrete file outputs and directives — harness/score.sh, lint.sh, probe.sh, eval/dev vs eval/holdout, 'output VOID: constraint violation', 'keyword list ≤ 20 entries' — give specific actionable guidance; as an instruction-only skill, executable code is not required per the rubric's code_vs_instruction note.	3 / 3
Workflow Clarity	Phases 0–9 plus Patch mode are explicitly sequenced, with validation checkpoints in Phase 6 (five concrete self-verification checks), a red-team feedback loop in Phase 7, and a stop criterion of three clean simulations.	3 / 3
Progressive Disclosure	The overview body points one level deep to real, verified bundle files — references/cheat-museum.md, references/log-template.md, references/goal-template.md — keeping the main skill lean with well-signaled navigation.	3 / 3
	Total	12 / 12 Passed

Description

100%

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description is specific, trigger-rich, and complete: it states what the skill does, when to use it, and when to re-invoke it, all in a distinctive niche. It would not score higher because the rubric tops out at 3; it is not a 2 because the what/when pair is explicit rather than implied.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'Observes the existing environment, interrogates the task, ingests or generates the spec, builds a blinded eval, generates and verifies the harness, red-teams the target for cheats, and emits goal.md' — matching the multiple-specific-actions anchor.	3 / 3
Completeness	Explicitly answers both what (design a loss function and harness, enumerating the steps) and when ('Use when the user wants to set up...'), plus a re-invocation condition, satisfying the explicit-trigger anchor.	3 / 3
Trigger Term Quality	Natural trigger phrasing users of the system would say — 'autonomous optimization loop', 'distill a product from public artifacts', 'turn a spec into an optimization target', 'designs a /goal' — gives good coverage rather than jargon.	3 / 3
Distinctiveness Conflict Risk	A narrow niche — loss-function development for /goal runs — with distinct /goal-anchored triggers unlikely to fire for unrelated skills.	3 / 3
	Total	12 / 12 Passed

Validation

100%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: elvisun/loss-function-development
Commit: abf2661

Reviewed: 1 day ago

Table of Contents

Discovery Implementation Validation

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.