agentleFS
aws-solutionsaws-solutions/konductor/skills/workspace-skills/SKILL.md

workspace-skills

Skill12 starsChanged 54 days agoApache-2.0View on GitHub

Use when told "save this as a skill", "remember how to do this" or "make this repeatable", or after trial-and-error, a correction or an unexpected outcome. Also use to update a saved workspace procedure. Captures the reusable procedure as a project-specific workspace skill, with proactive capture, dedup, and provenance protection.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When a quality signal fires, propose the capture with a single lightweight confirm line — keep it low-friction:

    line 92
  2. Very specific:

    After the user confirms, write the file and notify: "Captured skill <name> — <desc>."

    line 97
  3. Distinctive:

    Write the offer in third person, stating WHAT the skill does AND WHEN to use it, including trigger phrases the user might say.

    line 100
  4. Very specific:

    Check the ws-* workspace skills already visible in your context.

    line 112
  5. Very specific:

    Every agent-created skill MUST carry origin: agent-created in its frontmatter.

    line 134
  6. Very specific:

    If you cannot determine provenance, treat the entry as user-authored.

    line 146
  7. Very specific:

    When a user-authored skill needs updating, create a new agent-authored skill with related: [original-name] rather than modifying the user's …

    line 147
  8. Distinctive:

    All four of the following.

    line 155

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Workspace Skills

Routing note

Reusable multi-step procedures and workflows belong here. Declarative facts, preferences, decisions, and conventions belong in persistent-memory. When unsure, apply this rule:

Captured knowledge is...Route to
A reusable procedure or workflowworkspace-skills
A fact, preference, or conventionpersistent-memory
A one-time observation (not durable)Neither — discard

When to capture a skill (proactive — do not wait to be asked)

Primary triggers (high-quality signal — act on any one)

  • Trial-and-error that succeeded: you tried ≥2 different approaches before finding what worked
  • Course correction mid-task: you changed approach partway through because the first path failed
  • User corrected your approach and the corrected approach worked
  • Unexpected outcome requiring investigation: a non-obvious finding that required diagnosis

Secondary trigger (weak signal — use only when all above are absent)

  • The task took 5+ tool calls with a coherent procedural arc (mechanical repetition of the same tool does NOT qualify)

Suppress (do not capture)

  • Mechanical repetition of the same tool call N times with no variation
  • Simple one-off answers with no reusable structure
  • Anything already covered by an existing skill (patch instead)

Phase-1 Decision Flow

Agent completes a task step
        |
        v
Any Primary trigger fired?
        |
   YES -+- NO -> Any Secondary trigger with coherent arc?
        |                |
        |           YES -+- NO -> continue (nothing to capture)
        |                |
        v                v
Classify: Fact -> persistent-memory | Procedure -> workspace-skills
        |
        v
Scan existing skills for overlap (see Scan-Before-Create)
        |
  Overlap? --YES--> Check origin marker
        |                |
        NO               | [origin:agent-created] --> PATCH existing skill
        |                | [origin:user-authored] or untagged --> CREATE new
        v                v                                       + note related:
  CREATE new          (PATCH or CREATE)
  origin: agent-created    |
        |                  |
        +------------------+
        |
        v
  Lightweight confirm: "Capturing skill: <name> — <one-line desc>. OK?"
        |
        v
  Write skill file
        |
        v
  Notify: "Captured skill: <name> — <desc>"

Offer, then confirm

When a quality signal fires, propose the capture with a single lightweight confirm line — keep it low-friction:

"Capturing skill: <name> — <one-line desc>. OK?"

After the user confirms, write the file and notify: "Captured skill <name> — <desc>."

Write the offer in third person, stating WHAT the skill does AND WHEN to use it, including trigger phrases the user might say. The provenance markers and immutability gate (see Provenance Markers below) are the safety net — they protect user-authored content regardless of how capture is initiated.


Scan-Before-Create (dedup gate — mandatory)

Before creating any new skill entry, you MUST check for overlap.

Check the ws-* workspace skills already visible in your context. On Kiro CLI they are loaded at session start via the skill://.kiro/skills/ws-*/SKILL.md resource glob (re-resolved each session). On Claude Code, list .claude/skills/ws-*. Each workspace skill's name and description appear in your context — scan those for overlap before writing anything.

Decision rules:

  • Existing skill covers the same use case → PATCH it (only if origin: agent-created; see Provenance below)
  • Overlap is partial and use cases differ → CREATE new + add related: [existing-name] in frontmatter
  • Unsure → default to patching ONLY if origin: agent-created; otherwise CREATE new + related: [existing-name] (so user-authored skills are never patched under uncertainty)

Duplicates dilute discovery and degrade retrieval quality.


Provenance Markers (user entries are immutable to agents)

Every agent-created skill MUST carry origin: agent-created in its frontmatter. When the user explicitly dictates a skill, write origin: user-authored.

Provenance gate — check BEFORE any modification

Marker in frontmatterAgent may PATCH?
origin: agent-createdYes
origin: user-authoredNever
No origin field (untagged / legacy)Never

If you cannot determine provenance, treat the entry as user-authored. When a user-authored skill needs updating, create a new agent-authored skill with related: [original-name] rather than modifying the user's entry.


What a worth-saving skill contains

All four of the following. If any is missing, the workflow is not stable enough yet — keep iterating, do not save.

  1. Trigger conditions — when to use this skill (be specific)
  2. Numbered steps with exact commands or tool calls
  3. Pitfalls section — what goes wrong and how to avoid it
  4. Verification steps — how to confirm the procedure succeeded

How to Create a Skill

  1. Scan for overlap using the ws-* workspace skills already in context (see Scan-Before-Create above) — the scan result determines whether to PATCH an existing skill or CREATE a new one before offering.
  2. Offer with a lightweight confirm line: "Capturing skill: <name> — <one-line desc>. OK?" Keep it low-friction.
  3. Write to the appropriate path for the current runtime after confirmation:

    • .kiro/skills/ws-<name>/SKILL.md (Kiro CLI)
    • .claude/skills/ws-<name>/SKILL.md (Claude Code)

    The skill is local; whether to commit it is the developer's choice.

  4. Notify the user: "Captured skill <name> — <desc>."

The ws- prefix is mandatory for both runtimes — it prevents name collisions with package-provided skills.


Frontmatter Schema

---
name: ws-<name>         # kebab-case, ws- prefix required
description: <one sentence>  # what procedure this skill captures
origin: agent-created   # or: user-authored
version: 1              # integer, increment on every edit
last_used: YYYY-MM-DD   # updated after applying the skill
related: []             # optional: names of related/overlapping skills
---

Example Skill Body

---
name: ws-dynamodb-pagination
description: >
  How to implement DynamoDB pagination with LastEvaluatedKey
  in this project.
origin: agent-created
version: 1
last_used: 2026-05-21
related: []
---

# DynamoDB Pagination

## When to use

- Implementing a list/query endpoint that returns more items than fit in a single response
- Adding cursor-based pagination to an existing DynamoDB `QueryCommand` or `ScanCommand`
- Debugging incomplete result sets that silently drop items due to missing `LastEvaluatedKey` handling

## Pattern

Always use `ExclusiveStartKey` + `LastEvaluatedKey`. Never use `Limit`
alone as a pagination signal.

## Steps

1. Accept optional `nextToken` query param (base64-encoded key)
2. Decode token -> pass as `ExclusiveStartKey` to `QueryCommand`
3. If response has `LastEvaluatedKey`, encode -> return as `nextToken`
4. Return `{ items, nextToken }` -- omit `nextToken` on last page

## Pitfalls

- Using `Limit` as a stop signal returns partial pages when items are
  filtered post-scan
- Forgetting to base64-encode the token before returning to the client

## Verification

- Paginate through a table with 3+ pages; confirm all items returned
- Request with an invalid token returns a clear error, not empty page

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.