continuous-improvement
Skill88 starsChanged 3 days agoMITView on GitHub
Protocol for updating skills based on user feedback and root-cause analysis to prevent recurring mistakes. Use this skill when deciding HOW and WHEN to evaluate or update other agent skills.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- BurakD/ai-engineering-harnesstheir own version, first seen Nov 1, 2026
- naimkatiman/continuous-improvementtheir own version, first seen Oct 13, 2026
- Abdox-menara/guard-skillstheir own version, first seen Aug 24, 2026
- kaka-ruto/skillstheir own version, first seen Apr 23, 2026
- muyen/vibe-to-prodtheir own version, first seen Jan 3, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- muyen/vibe-to-prod★ 34same name
- naimkatiman/continuous-improvement★ 9same name
- kaka-ruto/skills★ 4same name
- BurakD/ai-engineering-harness★ 1same name
- Abdox-menara/guard-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository52 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- alchemy-orchestrateSkill
- alchemySkill
- architecture-as-code-javascriptSkill
- architecture-as-code-pythonSkill
- architecture-as-codeSkill
- architecture-guidelinesSkill
- bring-downSkill
- ci-cd-reliability-architectureSkill
- defect-shift-leftSkill
- dependency-lifecycleSkill
- evolutionary-database-designSkill
- functionality-complexity-tradeoffSkill
- gate-lifecycleSkill
- implementation-readinessSkill
- infrastructure-fitSkill
- morphogenetic-architectureSkill
- observability-designSkill
- push-outSkill
- requirements-groundingSkill
- requirements-topologySkill
- requirements-traceabilitySkill
- structural-simplificationSkill
- system-optimizationSkill
- test-strategySkill
- zero-copy-requirementsSkill
- alchemy-orchestrateSkill
- alchemySkill
- architecture-as-code-javascriptSkill
- architecture-as-code-pythonSkill
- architecture-as-codeSkill
- architecture-guidelinesSkill
- bring-downSkill
- ci-cd-reliability-architectureSkill
- continuous-improvementSkill
- defect-shift-leftSkill
- dependency-lifecycleSkill
- evolutionary-database-designSkill
- functionality-complexity-tradeoffSkill
- gate-lifecycleSkill
- implementation-readinessSkill
- infrastructure-fitSkill
- morphogenetic-architectureSkill
- observability-designSkill
- push-outSkill
- requirements-groundingSkill
- requirements-topologySkill
- requirements-traceabilitySkill
- structural-simplificationSkill
- system-optimizationSkill
- test-strategySkill
- zero-copy-requirementsSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
15 rules. 0 are common, 12 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 4 edited versions we read keep these.
12
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 56
When the subject is the skills themselves, optimize in this order:
- Distinctive: line 62
Check whether validation can catch it: mirror sync, frontmatter, primer backlinks, output contract presence, owner/routing conflicts, or …
- Very specific: line 67
Keep .agents/skills, .claude/skills, and the matching .documentation/READ-*.md primer aligned in the same change.
- Very specific: line 70
Run python scripts/validate-skills.py.
- Distinctive: line 75
Use this sequence when learning originates in a consumer project:
- Distinctive: line 81
Update the one canonical generic owner in the skill library.
- Distinctive: line 83
Update mirrors, primers, indexes, root routing guidance, and validation in the same library change.
- Distinctive: line 89
Do not overwrite a generic library from a consumer tree or treat byte-identical vendored files as locally owned.
- Very specific: line 124
When generating the actual update to the relevant SKILL.md file:
- Distinctive: line 126
Actionable Constraints: Ensure new rules are testable, direct, scoped, and explicit about exceptions.
- Distinctive: line 128
Review Sibling Skills: Check existing skills to ensure you are not creating an ownership conflict or duplicating the rule's primary …
- Distinctive: line 133
Maintain Mirrors and Validate: keep mirrors, primers, and the index aligned as Optimization Mode steps 4–5 require.
- Distinctive: line 139
Ensure the learning "sticks":
- Distinctive: line 141
Verification: If applicable, add a unit test, lint rule, schema check, or perform an audit showing the previous mistake is now caught …
- Distinctive: line 155
When applying this skill, emit:
“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Continuous Improvement Protocol (Meta-Learning)
This protocol governs the agent's meta-learning process (Kaizen). It ensures that SKILL files evolve organically, preventing recurring errors through systemic, root-cause adjustments, without growing bloated.
Core Directives
- Automation Before Prose: Before adding a manual instruction to a SKILL document, check whether the lesson can be encoded as a test, linter, schema, type, build gate, or template. Observe a new check failing without the rule before calling it enforced; a check never seen to fail is a hypothesis, not a safeguard. If automation is infeasible or too costly, record why and add the smallest useful prose rule.
- Density Over Volume: Prefer replacing, merging, or tightening existing rules before adding new ones. Prune only rules proven obsolete, contradicted, or owned elsewhere.
- Root-Cause Focus: Do not preserve a symptom as the rule. Trace the error to its root cause (e.g., misunderstood framework, missing boundary rule) and fix the most upstream reusable guidance.
- Single Owner, Explicit References: Keep each rule's primary authority in one SKILL. Other SKILLs may mention it briefly as a cross-reference when routing users to the owner would prevent misuse.
- Operationalize Before Broadening: For skill-library updates, add or extend validation, templates, or primer checks before broad prose changes. If prose is still needed, keep the rule at the owning SKILL and route siblings to it.
- Promote Before Repinning: When a consumer project proves a reusable lesson, update the canonical generic skill library first. Publish that revision, then repin the consumer; keep project facts in a local overlay.
When to use this skill
- Triggered by user corrections, systemic failures, or regression triggers that are likely to recur beyond the current conversation (full catalogue in §1).
- Triggered when the user asks to improve, optimize, prune, or audit the skill library itself.
- Out of scope: This skill dictates the triggers, analysis, and mindset
for updating SKILLs. It does NOT dictate the formatting/layout of the SKILLs
(follow your project's skill-authoring conventions) nor does it define the
architectural code rules themselves (see
architecture-guidelines).
Skill-Library Optimization Mode
When the subject is the skills themselves, optimize in this order:
- Name the observed failure the change prevents — a correction, transcript, incident, or scenario. Preferred wording, or a request to extend or complete a skill, is not one; without it, return a proposal with an M verdict instead of editing.
- Check whether validation can catch it: mirror sync, frontmatter, primer backlinks, output contract presence, owner/routing conflicts, or template shape.
- Edit the owning
SKILL.mdonly. Sibling skills get cross-references unless they must make a routing decision. - Keep
.agents/skills,.claude/skills, and the matching.documentation/READ-*.mdprimer aligned in the same change. If the skill's public role or trigger changed, update theREADME.mdskill index too. - Run
python scripts/validate-skills.py. If no mechanical check can cover the lesson, record why in the output contract.
Consumer-to-Library Promotion
Use this sequence when learning originates in a consumer project:
- Capture the concrete failure, correction, or repeated successful pattern and its verification evidence in the consumer.
- Extract the smallest behavior that applies without the project name, private paths, roles, domain taxonomy, local commands, or provider assumptions.
- Update the one canonical generic owner in the skill library. Add a skill only when the capability has a distinct trigger, lifecycle, and output contract.
- Update mirrors, primers, indexes, root routing guidance, and validation in the same library change.
- Validate and publish the library revision.
- Repin or reinstall that revision in the consumer, then retain only its project profile and other domain-specific skills locally.
Do not overwrite a generic library from a consumer tree or treat byte-identical vendored files as locally owned. If a local emergency patch is unavoidable, mark it as temporary divergence with an upstream owner and remove it when the published fix is repinned.
1. Triggers for Learning
Initiate this protocol to examine and update existing SKILL.md files when
encountering:
| Trigger Type | Scenario |
| Correction | User explicitly corrects an architectural pattern, approach, or rule. |
| Regression | A fix for one issue breaks another, indicating an undocumented system boundary. |
| New Pattern | A new, validated structural standard is successfully introduced to the codebase. |
| Systemic Failure | The same class of anti-pattern or fragile workaround is attempted repeatedly, or once with high blast radius. |
| Process Break | Recurring CI/CD failures or linter bottlenecks indicating a broken foundational rule. |
| Skill Optimization | User asks to optimize, prune, audit, or increase efficiency of the skill library itself. |
2. Analyze Root Cause
Determine the underlying reason before writing a new rule. Ask:
- Missing/Ambiguous: Was the requirement undocumented, too broad, too absolute, or written with the wrong force?
- Conflict: Did two SKILL rules contradict each other?
- Ignored Rule: Was a rule clear but ignored due to lack of visibility in the file? Improve placement or wording before promoting it to a warning block.
- Technical Constraint: Was the failure due to a misunderstood platform or framework behavior?
3. The Update Execution
When generating the actual update to the relevant SKILL.md file:
- Actionable Constraints: Ensure new rules are testable, direct, scoped, and explicit about exceptions.
- Review Sibling Skills: Check existing skills to ensure you are not creating an ownership conflict or duplicating the rule's primary authority.
- Eliminate Outdated Logic: Replace or delete previous rules only when they contradict, overlap with, or are made obsolete by the new learning. Preserve useful exceptions and migration notes.
- Maintain Mirrors and Validate: keep mirrors, primers, and the index aligned as Optimization Mode steps 4–5 require; extend the validator when a drift pattern can be detected mechanically.
4. Verification & Notification
Ensure the learning "sticks":
- Verification: If applicable, add a unit test, lint rule, schema check, or
perform an audit showing the previous mistake is now caught earlier or made
less likely. Do not claim structural impossibility unless an enforced gate
guarantees it. For skill-library changes, run
python scripts/validate-skills.py. A changed decision rule is a behavioral hypothesis: record the originating miss as a scenario that fails before the change and passes after, and apply the new rule to the change itself before calling it ready. - Notify: Conclude the improvement sequence with a concise summary back to
the user:
"Updated [Skill/Test] to prevent [issue] by mandating [new practice]."
5. Output Contract
When applying this skill, emit:
Subject: <skill / test / linter / schema / template>
Trigger: <correction | regression | new pattern | systemic failure | process break | skill optimization>
Root cause: <missing/ambiguous | conflict | ignored rule | technical constraint | ownership gap>
Owner: <skill/test/linter/schema/template that should own the fix>
Automation: <implemented | infeasible | deferred> - <reason>
Decision: <replace | add | delete | cross-reference | no change>
Verification: <checks performed, and the observed failure without the rule>
Next action: <edit made, check added, follow-up owner, or stop>
Residual risk: <what still depends on judgment or future enforcement>6. See also
architecture-guidelines— first-principles rules that updates from this skill propagate back into.defect-shift-left— when a recurring correction can be encoded as a Stage 0–5 gate instead of a SKILL rule (Directive 1).gate-lifecycle— keeps a lesson's gate proven after it lands.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

