improvement-loop

Use after resolving a bug, failed task, or unexpected agent behavior to improve the pipeline skills, agents, hooks, or scripts that contributed to the problem. Also proactively suggest improvements when recurring patterns or inefficiencies are observed.

Install with Tessl CLI

npx tessl i github:aaddrick/claude-desktop-debian --skill improvement-loop

What are skills?

1.25x

Quality

72%

Does it follow best practices?

Impact

93%

1.25x

Average score across 3 eval scenarios

Optimize this skill with Tessl

npx tessl skill review --optimize ./.claude/skills/improvement-loop/SKILL.md

SKILL.md

Review

Evals

Evaluation results

100%

31%

Pipeline Improvement After Code Review Bug

Improvement cycle workflow

Criteria

Without context

With context

Gate check documented

88%

100%

Improvement classified

75%

100%

Minimal change only

90%

100%

Anti-pattern added

100%

No speculative additions

100%

Verification noted

100%

Commit format: improve prefix

100%

Commit format: file and reason

62%

100%

User communication: Updated [file]

50%

100%

User communication: prevents [problem]

50%

100%

Direct edit (not new skill)

100%

Only one file changed

100%

Without context: $0.3854 · 1m 35s · 26 turns · 31 in / 4,905 out tokens

With context: $0.4150 · 1m 33s · 21 turns · 24 in / 4,866 out tokens

100%

Data Validation Utility with Pipeline Observation

Proactive improvement detection

Criteria

Without context

With context

No pipeline file modified

100%

improvement_suggestion.md exists

100%

Specific problem described

100%

Pipeline file named

100%

Task context mentioned

100%

Asks rather than acts

75%

100%

Change scope described

100%

I noticed [X] while working on [Y] opener

100%

Asks for permission explicitly

25%

100%

Main task complete

100%

Without context: $0.3216 · 1m 19s · 19 turns · 25 in / 4,240 out tokens

With context: $0.3967 · 1m 23s · 20 turns · 25 in / 4,511 out tokens

80%

16%

Pipeline Improvement Proposal After Multi-Issue Session

Improvement batching and classification

Criteria

Without context

With context

All three improvements listed

100%

Numbered format

100%

Option A: batch

62%

100%

Option B: individual

37%

100%

Option C: skip

100%

Incident 1 classified correctly

100%

Incident 2 classified correctly

100%

Incident 3 classified correctly

100%

Incident 2 routing: bash-script-craftsman

Incident 3 routing: cc-orchestration-writer

Task context per item

100%

Without context: $0.1551 · 1m · 9 turns · 14 in / 2,288 out tokens

With context: $0.2891 · 1m 11s · 14 turns · 17 in / 3,357 out tokens

Evaluated: 4 days ago
Agent: Claude Code
Model: Claude Sonnet 4.6

Table of Contents

Pipeline Improvement After Code Review Bug Data Validation Utility with Pipeline Observation Pipeline Improvement Proposal After Multi-Issue Session

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.