CtrlK
BlogDocsLog inGet started
Tessl Logo

cekura-self-improving-agent

Use to close the loop on agent quality — turn a failure signal into a verified fix. Triggers: "improve my agent", "self-improving agent", "auto-tune / iterate on my prompt", "fix my agent from test results", "optimize my prompt based on failures", "rewrite my prompt". ALSO for production-call bug fixing: "fix this prod call issue", "debug and fix call ID", "reproduce this production bug". Works across VAPI, Retell, ElevenLabs, Bland, and self-hosted agents, and across three fix surfaces — prompt, tool config, and (self-hosted) owned source code, including infra-flavored / forked-SDK bugs, which are reproduced and validated on Cekura (never a code test).

70

Quality

86%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

High

Do not use without reviewing

SKILL.md
Quality
Evals
Security

Quality

Content

77%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.

A crisply sequenced orchestration playbook with strong validation gates and feedback loops, but its progressive-disclosure design is undercut by a bundle that omits the phases/ and providers/ files the body links to while leaving the present references/ files unlinked.

Suggestions

Ship the referenced phase and provider files (phases/*.md, providers/*/overview.md) so the body's inline links resolve, or remove the dead links and inline the essentials.

Link the existing references/*.md files (phase-2-failure-collection.md, phase-3-diagnosis.md, dynamic-variables-debugging.md) from the body instead of leaving them orphaned.

Expand the bare acronyms (COLLECT.6, REPRO.3e, FIX.1) on first use, or consolidate them into a single glossary line, to remove the one conciseness tax without adding length.

DimensionReasoningScore

Conciseness

Lean, jargon-dense prose that assumes Claude's competence ("Phases run strictly in sequence ... each consumes the previous phase's output as a hard pre-condition") with no concept over-explanation, matching the efficient-with-minor-trim anchor; not a 5 due to unexpanded acronyms (COLLECT.6, REPRO.3e, FIX.1) and some redundancy with the frontmatter.

4 / 5

Actionability

Concrete orchestration guidance with explicit gates ("it must fail ≥ M of N before any edit"), parameterized defaults/ranges ("dataset_size (default 8, range 5–10)", "repro_threshold (default ⌈runs/2⌉"), and named tool calls, matching the mostly-executable-with-minor-gaps anchor; not a 5 because the executable specifics are delegated to referenced phase/provider files that are not present in the bundle.

4 / 5

Workflow Clarity

Clear 9-phase sequence with an explicit loop re-entry point ("Loop point: Eval → Collect"), must-fail-first and must-pass stochastic gates, and feedback loops (Sync→Apply rollback, Regression→Collect, "Non-zero redeploy exit halts"), matching the anchor with explicit validation steps and error-recovery loops for batch/destructive operations.

5 / 5

Progressive Disclosure

The overview is well-organized with a Files manifest and one-level-deep inline links, but scored against the actual bundle: the referenced phases/ and providers/ trees do not exist on disk (only references/ is present) and the three references/*.md files are not linked from the body, so navigation is effectively broken and existing references are orphaned.

3 / 5

Total

16

/

20

Passed

Description

95%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.

A strong, specific description with comprehensive natural triggers that explicitly cover both what and when across a well-scoped niche. Minor tightness lost only because capabilities are framed more than exhaustively enumerated.

DimensionReasoningScore

Specificity

Names the domain and several concrete actions ("turn a failure signal into a verified fix", "three fix surfaces — prompt, tool config, and ... owned source code", "reproduced and validated on Cekura"), matching the anchor that lists several specific actions with minor gaps; it is not a 5 because it frames capability rather than comprehensively enumerating many distinct concrete actions.

4 / 5

Completeness

Explicitly answers both what ("turn a failure signal into a verified fix ... across three fix surfaces ... reproduced and validated on Cekura") and when ("Triggers: ...", "ALSO for production-call bug fixing: ...") with concrete trigger phrases, matching the top anchor.

5 / 5

Trigger Term Quality

Comprehensive natural trigger phrases with synonyms ("improve my agent", "self-improving agent", "auto-tune / iterate on my prompt", "optimize my prompt based on failures", "rewrite my prompt") plus production-bug variants ("fix this prod call issue", "debug and fix call ID", "reproduce this production bug"), matching the comprehensive-synonyms anchor.

5 / 5

Distinctiveness Conflict Risk

Clear niche scoped to Cekura agent-quality loops over named providers (VAPI, Retell, ElevenLabs, Bland, self-hosted) with distinct triggers and an explicit "never a code test" boundary, giving minimal conflict risk.

5 / 5

Total

19

/

20

Passed

Validation

93%

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

Validation15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

relative_links

Relative link issues: 16 missing, 8 deeper-than-1-level

Warning

Total

15

/

16

Passed

Repository
cekura-ai/cekura-skills
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.