CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

self-evolving-single-agent

Use when generating a single installable agent that should keep learning, track sources, refresh research, propose repairs, or improve itself over time without becoming a multi-agent team.

Quality

68%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Medium

Suggest reviewing before use

Fix and improve this skill with Tessl

tessl review fix ./claude/plugins/agentlas-core-engine-meta-agent/runtime-support/skills/self-evolving-single-agent/SKILL.md

The canonical home for this skill is self-evolving-single-agent in agentlas-ai/Agentlas-OS

SKILL.md
Quality
Evals
Security

Quality

Content

57%Weight 40%Scale 1-5

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

The skill body is compact and well-sectioned with a distinctive approval-gate design, but it operates mostly at the naming level: files and sections are enumerated without content templates, executable steps, or a verification procedure, and one referenced contract file does not exist in the bundle. It reads more like a checklist skeleton than an actionable procedure.

Suggestions

Resolve the `contracts/builder-interview-research-gate.md` reference — either ship it in a bundle directory or state that it is created as part of the output package — and signal which of the other named files are inputs vs. outputs.

Add concrete templates or field-level guidance for the key artifacts (e.g., the structure of `.agentlas/memory-map.json` and `capability-eval-plan.json`, and what the `verification` output must contain and how to produce it).

Spell out the workflow's validation checkpoints: what the interview/research gate must produce before generation starts, and how to verify the generated package before returning it.

DimensionReasoningScore

Conciseness

The body is dense and lean — numbered steps, exact file paths, no concept explanations Claude already knows, no padding (e.g., 'Never treat a summary, low-risk label, Curator verdict, shell flag, or legacy chip as approval' is compact and specific). It sits at anchor 4 rather than 5 only because a few spots could still be tightened, such as the overlapping memory rules in step 3 ('project memory written only through Memory Tickets' plus 'Memory Events and Memory Tickets for durable updates').

4 / 5

Actionability

Guidance is partially concrete — specific file names (`.agentlas/memory-map.json`, `docs/research-sources.md`) and a numbered procedure — but there are no commands, examples, or templates, and key details are missing: what goes inside memory-map.json, how the worker prompt should be structured, and what the verification output must contain. The referenced `contracts/builder-interview-research-gate.md` is not in any bundle directory, so the instruction cannot be executed as written; this matches 'some concrete guidance but incomplete; missing key details'.

3 / 5

Workflow Clarity

The seven steps give a clear sequence, and step 6 plus the File revision approval section provide an explicit approval checkpoint. However, the required `verification` output is named with no process behind it (no validate/fix/retry loop for the generated package), and the interview gate's exit criteria are only implied — fitting anchor 3's 'sequence present but checkpoints missing or implicit' rather than anchor 4's 'most checkpoints present'.

3 / 5

Progressive Disclosure

The body has decent section structure (Procedure / Output / File revision approval) and no bundle files exist to organize, but the reference to `contracts/builder-interview-research-gate.md` is neither a bundled file nor clearly marked as an artifact to create — it is an unsignaled, unresolved path. Per the rubric guideline to score against the actual (empty) bundle structure, this lands at anchor 3 ('references present but not clearly signaled') rather than 4.

3 / 5

Total

13

/

20

Passed

Description

62%Weight 40%Scale 1-5

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 has a clear, distinctive niche and an explicit 'Use when' trigger, but it describes the artifact's desired traits more than the skill's own actions and misses common user phrasings like 'self-improving' or 'self-evolving'. It reads as a trigger-first description with the 'what' implicitly folded in.

Suggestions

State the skill's concrete actions up front (e.g., 'Generate a single installable agent package with memory architecture, research refresh loop, and approval-gated self-evolution.') before the 'Use when' clause.

Add natural synonyms users would actually say — 'self-improving', 'self-evolving', 'autonomous agent', 'agent that learns from experience' — to the trigger phrasing.

Make the 'when' clause more specific about the situations that call for this skill (e.g., 'when the user asks for one agent that improves itself over time instead of a multi-agent team').

DimensionReasoningScore

Specificity

The description names one concrete action ("generating a single installable agent") and then lists attributes of the artifact ("keep learning, track sources, refresh research, propose repairs, or improve itself over time") rather than actions the skill itself performs. It matches 'names domain and 1-2 concrete actions, but not comprehensive' — above score 2 because the generating action is concrete, below score 4 because no several specific actions are enumerated.

3 / 5

Completeness

The 'when' is explicit ("Use when generating a single installable agent that...") and the 'what' is embedded in the same clause (generate a single agent with learning/source-tracking/self-repair behaviors). Both are present, so it clears the anchor-4 bar, but the 'what' is folded into the trigger clause rather than stated independently with concrete trigger phrases, keeping it below anchor 5.

4 / 5

Trigger Term Quality

Some natural phrases are present ("keep learning", "improve itself over time", "agent", "multi-agent team"), but common variations users would say — "self-improving", "self-evolving", "autonomous", "learns from experience" — are absent, and "track sources", "propose repairs", "installable" lean toward internal jargon. Fits anchor 3 ('some relevant keywords but missing common variations or synonyms') better than anchor 4's 'good keyword coverage'.

3 / 5

Distinctiveness Conflict Risk

The niche is well carved out — a single worker that self-improves, explicitly contrasted with "without becoming a multi-agent team" — giving minimal overlap with team-orchestration skills. It stays at 4 rather than 5 because it could still trigger alongside generic agent-generation/builder skills that share the "generating an agent" trigger.

4 / 5

Total

14

/

20

Passed

Validation

100%

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

agentlas-ai/Agentlas-OS

Path

claude/plugins/agentlas-core-engine-meta-agent/runtime-support/skills/self-evolving-single-agent/SKILL.md

Reviewed

Table of Contents

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.