⏳ This skill is pending AI review.
Scores will appear once the review pipeline completes.
tier-1-2-3-skill-system
Use when creating, editing, reviewing, packaging, or organizing AI-agent skills and deciding whether a skill should be text-only, script-backed, or backed by an ML/specialist pipeline.
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
Not yet listed on ClawHub or SkillsMP
// README
Tier 1-2-3 Skill System
A simple way to decide how heavy an AI-agent skill should be.
A skill is reusable knowledge for an AI agent: when a certain kind of task comes up, the skill tells the agent how to handle it well.
The main idea: do not make every skill a giant prompt, and do not overbuild every workflow into software. Use the lightest version that works.
The three tiers
Tier 1: Text-only
Use plain instructions when the agent mainly needs judgment.
Good for:
- Writing style
- Review checklists
- Research strategy
- Design principles
- Policy or safety rules
Tier 1 teaches the agent what to do and how to think.
Tier 2: Text + script
Add scripts when the agent keeps doing the same mechanical work.
Good for:
- Validating files
- Converting formats
- Packaging projects
- Generating reports
- Running repeatable checks
Tier 2 gives the agent reliable machinery instead of making it rewrite glue code every time.
Tier 3: Text + script + ML pipeline
Add an ML model or specialist pipeline only when the agent is missing a real capability.
Good for:
- Audio detection
- Screenshot scoring
- Search reranking
- Classification
- Domain-specific detection or evaluation
Tier 3 gives the agent a new sense organ.
Rule of thumb
Start with Tier 1.
Move to Tier 2 when the work becomes repetitive and mechanical.
Move to Tier 3 only when the missing piece is measurable and needs perception, ranking, scoring, detection, or classification.
Why this matters
This keeps skills:
- Easier to read
- Easier to share
- Less bloated
- More reliable
- Less likely to make agents improvise fragile code
Do not overbuild. The best skill system is lean at the top, scripted where useful, and ML-backed only when the payoff is real.
For agents
SKILL.md contains the agent-facing version of this model.
Install directly from GitHub:
npx skills add H-Ali13381/tier-1-2-3-skill-system
The root SKILL.md is canonical. A mirror copy also lives under .agents/skills/tier-1-2-3-skill-system/ so cross-client skill scanners can discover it without breaking strict root skill validation.
Related skill
Use this with Recursive Agent Improvement:
Tier 1-2-3 decides whether a workflow should stay text-only, become script-backed, or escalate to an ML/specialist pipeline. Recursive Agent Improvement is the companion for the Tier 3 path: building the specialist tool, evaluator, verifier, reranker, classifier, detector, or training pipeline.
License
MIT
// HOW IT'S BUILT
// FILES
5 files shipped with this skill. Inline text is the catalog copy. The upstream link is the source repository.
.agents/skills/tier-1-2-3-skill-system/SKILL.md7.6 KB4d5a99507cb8upstream ↗
--- name: tier-1-2-3-skill-system description: Use when creating, editing, reviewing, packaging, or organizing AI-agent skills and deciding whether a skill should be text-only, script-backed, or backed by an ML/specialist pipeline. version: 1.0.0 author: Hasan Ali license: MIT metadata: tags: - ai-agents - skills - skill-design - automation - ml-pipeline related_skills: - recursive-agent-improvement related_repositories: - https://github.com/H-Ali13381/recursive-agent-improvement --- # Tier 1-2-3 Skill System ## Overview Use this skill to choose the lightest reliable shape for an AI-agent skill. A skill is reusable operational knowledge. It should reduce future steering, capture real pitfalls, and make repeatable work more reliable. It should not become a bloated prompt dump or a session diary. Core rule: start simple, then promote only when the workflow proves it needs more machinery. ## When to Use Use this when: - Creating a new agent skill. - Reviewing or cleaning up an existing skill. - Deciding whether a workflow belongs in plain instructions, scripts, or a specialist model. - A skill is becoming too long, vague, or hard to operate. - Agents keep rewriting the same helper code for the same task. - A task needs perception, scoring, ranking, detection, or classification that the base agent cannot do reliably. Do not use this to justify overbuilding. The default answer is the lowest tier that works. ## The Three Tiers | Tier | Shape | Use when | Upgrade trigger | |---|---|---|---| | 1. Text-only | `SKILL.md` | The value is judgment, policy, sequencing, taste, style, or a checklist. | Agents repeatedly rewrite glue code or make mechanical mistakes. | | 2. Text + script | `SKILL.md` + `scripts/` | The workflow has deterministic, repeatable mechanics: parse, validate, convert, audit, launch, package, benchmark, or export. | Scripts and prompts cannot provide the missing perception, ranking, scoring, detection, or domain inference. | | 3. Text + script + ML pipeline | `SKILL.md` + `scripts/` + training/eval/model artifacts | The agent needs a non-native capability: classifier, detector, reranker, scorer, evaluator, verifier, or specialist model. | The specialist is stable enough to package as a tool, MCP server, CLI, or standalone skill. | Short version: - Tier 1 teaches the agent what to do. - Tier 2 gives the agent deterministic machinery to do it. - Tier 3 gives the agent a new sense organ. ## Tier 1: Text-only Use Tier 1 when the hard part is judgment. Good fits: - Writing voice - Review philosophy - Research synthesis - Design principles - Safety rules - Operational checklists Keep the skill focused on: - Trigger conditions - Decision rules - Common mistakes - Verification checklist - One or two concrete examples Avoid adding scripts when the agent only needs better taste, sequencing, or policy. ## Tier 2: Text + Script Use Tier 2 when the skill would otherwise make agents keep inventing the same glue code. Good fits: - File validation - Format conversion - Skill packaging - Report generation - Project audits - Benchmark aggregation - Repeatable setup or launch commands Rules: - `SKILL.md` explains when, why, and how to call the script. - `scripts/` does the deterministic work. - Scripts should be idempotent where practical. - Scripts should expose flags or environment variables instead of requiring edits. - Scripts should verify success before exiting 0. - Scripts should avoid logging secrets. - Bulky rationale belongs in `references/`, not the main skill body. Promotion signal: if an agent writes the same helper script twice, make it a bundled script. ## Tier 3: Text + Script + ML Pipeline Use Tier 3 when the base model lacks a measurable capability and ordinary scripts cannot fill the gap. Good fits: - Wakeword/audio detection - Screenshot or layout scoring - UI/theme coherence judging - Search reranking - Patch-risk classification - Fact-consistency verification - Domain-specific extraction - Anomaly detection Rules: - Define inputs, outputs, metrics, and baselines before training. - Include hard negatives and failure cases. - Keep inference compact and machine-readable, preferably JSON. - Explain when to trust the specialist and when to abstain. - Explain how the specialist output changes the agent's next action. - Add an eval plan and model card before publishing a specialist model. Promotion signal: the missing piece is perception, ranking, scoring, detection, classification, or domain inference. ## Related Skill Use `recursive-agent-improvement` when the decision reaches Tier 3 and the agent needs to turn a measurable failure into a specialist tool, evaluator, reranker, classifier, detector, verifier, or ML training pipeline. This skill chooses the lightest reliable shape; `recursive-agent-improvement` builds the Tier 3 capability. Repo: https://github.com/H-Ali13381/recursive-agent-improvement ## Decision Checklist Before writing or changing a skill, ask: 1. Is this durable operational knowledge? 2. Will it reduce future steering? 3. Which tier is the lightest reliable fit? 4. Can this be consolidated with an existing skill? 5. Is repeated mechanical work better handled by a script? 6. Is the missing capability actually ML-shaped and measurable? 7. What verification proves the skill works? ## Example ```text Task: Package a repeated workflow as a public agent skill. Decision: - If the hard part is judgment, sequencing, policy, or style, keep it Tier 1 with only SKILL.md. - If agents keep rewriting the same validation, conversion, audit, launch, or packaging code, promote it to Tier 2 and add a `scripts/` directory. - If the workflow needs a measured classifier, detector, reranker, scorer, evaluator, verifier, or specialist model, promote it to Tier 3. Verification: - The selected tier is the lightest reliable shape. - The SKILL.md description says when to use the skill. - Any referenced scripts, references, assets, models, or eval docs exist. ``` ## Common Mistakes - Writing a giant prompt when a small checklist would do. - Creating many overlapping skills instead of one clear router skill. - Making agents rewrite deterministic code that should be scripted. - Adding ML because it sounds impressive, not because the task has a measurable capability gap. - Putting private paths, secrets, or session-specific history into public skill docs. - Treating a skill as a completed-work log instead of reusable operational knowledge. ## Error Handling - If the tier is unclear, choose the lower tier and write down the concrete upgrade signal. - If a Tier 1 skill keeps causing repeated mechanical mistakes, add or reference a tested script instead of expanding the prompt. - If a Tier 2 script fails, fix the script, expose clearer flags, and make the failure machine-readable before adding more prose. - If Tier 3 has no metrics, eval set, or abstention rule, do not publish the model-backed version yet. - If validation fails, fix frontmatter, broken file references, hardcoded paths, or missing supporting files before sharing. ## Verification Before publishing a skill shaped by this model: - Confirm the frontmatter has at least `name` and `description`. - Confirm the description says when to use the skill, not just what it contains. - Confirm the main `SKILL.md` is lean enough to scan. - Confirm any referenced scripts or files exist. - Confirm scripts have safe defaults and success/failure signals. - Confirm Tier 3 models have documented inputs, outputs, metrics, and limitations. Final rule: do not overbuild. The strongest skill system is lean at the top, mechanical where needed, and ML-backed only when the payoff is real.SKILL.md7.6 KB4d5a99507cb8upstream ↗
--- name: tier-1-2-3-skill-system description: Use when creating, editing, reviewing, packaging, or organizing AI-agent skills and deciding whether a skill should be text-only, script-backed, or backed by an ML/specialist pipeline. version: 1.0.0 author: Hasan Ali license: MIT metadata: tags: - ai-agents - skills - skill-design - automation - ml-pipeline related_skills: - recursive-agent-improvement related_repositories: - https://github.com/H-Ali13381/recursive-agent-improvement --- # Tier 1-2-3 Skill System ## Overview Use this skill to choose the lightest reliable shape for an AI-agent skill. A skill is reusable operational knowledge. It should reduce future steering, capture real pitfalls, and make repeatable work more reliable. It should not become a bloated prompt dump or a session diary. Core rule: start simple, then promote only when the workflow proves it needs more machinery. ## When to Use Use this when: - Creating a new agent skill. - Reviewing or cleaning up an existing skill. - Deciding whether a workflow belongs in plain instructions, scripts, or a specialist model. - A skill is becoming too long, vague, or hard to operate. - Agents keep rewriting the same helper code for the same task. - A task needs perception, scoring, ranking, detection, or classification that the base agent cannot do reliably. Do not use this to justify overbuilding. The default answer is the lowest tier that works. ## The Three Tiers | Tier | Shape | Use when | Upgrade trigger | |---|---|---|---| | 1. Text-only | `SKILL.md` | The value is judgment, policy, sequencing, taste, style, or a checklist. | Agents repeatedly rewrite glue code or make mechanical mistakes. | | 2. Text + script | `SKILL.md` + `scripts/` | The workflow has deterministic, repeatable mechanics: parse, validate, convert, audit, launch, package, benchmark, or export. | Scripts and prompts cannot provide the missing perception, ranking, scoring, detection, or domain inference. | | 3. Text + script + ML pipeline | `SKILL.md` + `scripts/` + training/eval/model artifacts | The agent needs a non-native capability: classifier, detector, reranker, scorer, evaluator, verifier, or specialist model. | The specialist is stable enough to package as a tool, MCP server, CLI, or standalone skill. | Short version: - Tier 1 teaches the agent what to do. - Tier 2 gives the agent deterministic machinery to do it. - Tier 3 gives the agent a new sense organ. ## Tier 1: Text-only Use Tier 1 when the hard part is judgment. Good fits: - Writing voice - Review philosophy - Research synthesis - Design principles - Safety rules - Operational checklists Keep the skill focused on: - Trigger conditions - Decision rules - Common mistakes - Verification checklist - One or two concrete examples Avoid adding scripts when the agent only needs better taste, sequencing, or policy. ## Tier 2: Text + Script Use Tier 2 when the skill would otherwise make agents keep inventing the same glue code. Good fits: - File validation - Format conversion - Skill packaging - Report generation - Project audits - Benchmark aggregation - Repeatable setup or launch commands Rules: - `SKILL.md` explains when, why, and how to call the script. - `scripts/` does the deterministic work. - Scripts should be idempotent where practical. - Scripts should expose flags or environment variables instead of requiring edits. - Scripts should verify success before exiting 0. - Scripts should avoid logging secrets. - Bulky rationale belongs in `references/`, not the main skill body. Promotion signal: if an agent writes the same helper script twice, make it a bundled script. ## Tier 3: Text + Script + ML Pipeline Use Tier 3 when the base model lacks a measurable capability and ordinary scripts cannot fill the gap. Good fits: - Wakeword/audio detection - Screenshot or layout scoring - UI/theme coherence judging - Search reranking - Patch-risk classification - Fact-consistency verification - Domain-specific extraction - Anomaly detection Rules: - Define inputs, outputs, metrics, and baselines before training. - Include hard negatives and failure cases. - Keep inference compact and machine-readable, preferably JSON. - Explain when to trust the specialist and when to abstain. - Explain how the specialist output changes the agent's next action. - Add an eval plan and model card before publishing a specialist model. Promotion signal: the missing piece is perception, ranking, scoring, detection, classification, or domain inference. ## Related Skill Use `recursive-agent-improvement` when the decision reaches Tier 3 and the agent needs to turn a measurable failure into a specialist tool, evaluator, reranker, classifier, detector, verifier, or ML training pipeline. This skill chooses the lightest reliable shape; `recursive-agent-improvement` builds the Tier 3 capability. Repo: https://github.com/H-Ali13381/recursive-agent-improvement ## Decision Checklist Before writing or changing a skill, ask: 1. Is this durable operational knowledge? 2. Will it reduce future steering? 3. Which tier is the lightest reliable fit? 4. Can this be consolidated with an existing skill? 5. Is repeated mechanical work better handled by a script? 6. Is the missing capability actually ML-shaped and measurable? 7. What verification proves the skill works? ## Example ```text Task: Package a repeated workflow as a public agent skill. Decision: - If the hard part is judgment, sequencing, policy, or style, keep it Tier 1 with only SKILL.md. - If agents keep rewriting the same validation, conversion, audit, launch, or packaging code, promote it to Tier 2 and add a `scripts/` directory. - If the workflow needs a measured classifier, detector, reranker, scorer, evaluator, verifier, or specialist model, promote it to Tier 3. Verification: - The selected tier is the lightest reliable shape. - The SKILL.md description says when to use the skill. - Any referenced scripts, references, assets, models, or eval docs exist. ``` ## Common Mistakes - Writing a giant prompt when a small checklist would do. - Creating many overlapping skills instead of one clear router skill. - Making agents rewrite deterministic code that should be scripted. - Adding ML because it sounds impressive, not because the task has a measurable capability gap. - Putting private paths, secrets, or session-specific history into public skill docs. - Treating a skill as a completed-work log instead of reusable operational knowledge. ## Error Handling - If the tier is unclear, choose the lower tier and write down the concrete upgrade signal. - If a Tier 1 skill keeps causing repeated mechanical mistakes, add or reference a tested script instead of expanding the prompt. - If a Tier 2 script fails, fix the script, expose clearer flags, and make the failure machine-readable before adding more prose. - If Tier 3 has no metrics, eval set, or abstention rule, do not publish the model-backed version yet. - If validation fails, fix frontmatter, broken file references, hardcoded paths, or missing supporting files before sharing. ## Verification Before publishing a skill shaped by this model: - Confirm the frontmatter has at least `name` and `description`. - Confirm the description says when to use the skill, not just what it contains. - Confirm the main `SKILL.md` is lean enough to scan. - Confirm any referenced scripts or files exist. - Confirm scripts have safe defaults and success/failure signals. - Confirm Tier 3 models have documented inputs, outputs, metrics, and limitations. Final rule: do not overbuild. The strongest skill system is lean at the top, mechanical where needed, and ML-backed only when the payoff is real.README.md2.5 KB8b2d09d8918bupstream ↗
# Tier 1-2-3 Skill System A simple way to decide how heavy an AI-agent skill should be. A skill is reusable knowledge for an AI agent: when a certain kind of task comes up, the skill tells the agent how to handle it well. The main idea: do not make every skill a giant prompt, and do not overbuild every workflow into software. Use the lightest version that works. ## The three tiers ### Tier 1: Text-only Use plain instructions when the agent mainly needs judgment. Good for: - Writing style - Review checklists - Research strategy - Design principles - Policy or safety rules Tier 1 teaches the agent what to do and how to think. ### Tier 2: Text + script Add scripts when the agent keeps doing the same mechanical work. Good for: - Validating files - Converting formats - Packaging projects - Generating reports - Running repeatable checks Tier 2 gives the agent reliable machinery instead of making it rewrite glue code every time. ### Tier 3: Text + script + ML pipeline Add an ML model or specialist pipeline only when the agent is missing a real capability. Good for: - Audio detection - Screenshot scoring - Search reranking - Classification - Domain-specific detection or evaluation Tier 3 gives the agent a new sense organ. ## Rule of thumb Start with Tier 1. Move to Tier 2 when the work becomes repetitive and mechanical. Move to Tier 3 only when the missing piece is measurable and needs perception, ranking, scoring, detection, or classification. ## Why this matters This keeps skills: - Easier to read - Easier to share - Less bloated - More reliable - Less likely to make agents improvise fragile code Do not overbuild. The best skill system is lean at the top, scripted where useful, and ML-backed only when the payoff is real. ## For agents `SKILL.md` contains the agent-facing version of this model. Install directly from GitHub: ```bash npx skills add H-Ali13381/tier-1-2-3-skill-system ``` The root `SKILL.md` is canonical. A mirror copy also lives under `.agents/skills/tier-1-2-3-skill-system/` so cross-client skill scanners can discover it without breaking strict root skill validation. ## Related skill Use this with Recursive Agent Improvement: - https://github.com/H-Ali13381/recursive-agent-improvement Tier 1-2-3 decides whether a workflow should stay text-only, become script-backed, or escalate to an ML/specialist pipeline. Recursive Agent Improvement is the companion for the Tier 3 path: building the specialist tool, evaluator, verifier, reranker, classifier, detector, or training pipeline. ## License MITAGENTS.md746 B6ca109817752upstream ↗
# Agent Notes This repository publishes one Agent Skill: `tier-1-2-3-skill-system`. Canonical files: - `SKILL.md` is the source of truth for the skill. - `.agents/skills/tier-1-2-3-skill-system/SKILL.md` is a cross-client discovery mirror. When editing the skill, update the root `SKILL.md` first, then copy it to the mirror path so both `SKILL.md` files stay identical. Related skill: - `recursive-agent-improvement`: https://github.com/H-Ali13381/recursive-agent-improvement Installability check: ```bash npx skills add H-Ali13381/tier-1-2-3-skill-system --list ``` Validation check: ```bash npx skill-tools check . ``` Do not add private paths, secrets, local debug notes, or session-specific history to this public skill artifact.LICENSE1.0 KB29654029e7feupstream ↗
MIT License Copyright (c) 2026 Hasan Ali Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Plain-text copy, no login: /skill-bundles/h-ali13381-tier-1-2-3-skill-system.md