All-my-ai-needs skill-promotion-and-dedup

Promote imported skills into first-class categories while preventing duplicate-name discovery conflicts. Includes backup, verification, and rollback.

install

source · Clone the upstream repo

git clone https://github.com/codingSamss/all-my-ai-needs

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/codingSamss/all-my-ai-needs "$T" && mkdir -p ~/.claude/skills && cp -r "$T/platforms/hermes/skills/autonomous-ai-agents/skill-promotion-and-dedup" ~/.claude/skills/codingsamss-all-my-ai-needs-skill-promotion-and-dedup && rm -rf "$T"

manifest: platforms/hermes/skills/autonomous-ai-agents/skill-promotion-and-dedup/SKILL.md

source content

Skill Promotion and Dedup (Hermes)

Use this when users import many custom skills (e.g. under

~/.hermes/skills/imported-codex

) and want better discoverability/reliability.

When to use

User says imported skills are hard to trigger/find.
Same skill names exist in multiple categories (imported + official), causing inconsistent selection.
User wants to keep imported originals, but use promoted copies as active skills.

Core idea

Promote imported skills to first-class categories (social-media, research, media, etc.).
Keep imported originals as archive.
Disable duplicate active entries in archive (avoid two
```
SKILL.md
```
files with same
```
name:
```
).

Procedure

Inventory and mapping

Enumerate source skills under

~/.hermes/skills/imported-codex/*/SKILL.md

Create an explicit mapping:
```
skill_name -> target_category
```
.
Do not proceed with unmapped skills.

Backup and promote

Create backup root:
```
~/.hermes/skill-promotions/<timestamp>/
```
.
For each skill:
- If target exists, backup existing target first.
- Copy imported skill folder to target category path.
Save a manifest:
```
promotion-manifest.json
```
with source/target/backups.

Configure defaults where needed

If there are competing skills for same domain, clarify routing in descriptions.
Example used successfully:
- ```
bird-twitter
```
  : default for read-only Twitter via browser login.
- ```
xitter
```
  : official API / write actions.

Deduplicate active discovery

In archive category (
```
imported-codex
```
), rename
```
SKILL.md
```
->
```
SKILL.imported.md
```
.
Keep scripts/assets/references intact.
Optionally add
```
imported-codex/DESCRIPTION.md
```
stating this is archive-only.
Important: if duplicate active definitions remain,
```
skills_list
```
may surface unstable/partial results (some promoted skills appear missing by category). Disable archive
```
SKILL.md
```
first, then re-check.

4b) Optional deletion mode (when user wants no archive)

Create a one-file archive first (safety):

tar -czf ~/.hermes/skill-promotions/archive-before-delete/imported-codex-<ts>.tgz -C ~/.hermes/skills imported-codex

Delete archive directory:
- ```
rm -rf ~/.hermes/skills/imported-codex
```
Verify:
- ```
skills_list(category='imported-codex')
```
  returns empty
If user asks for zero backup retention, delete the tarball too.

Verify

```
skills_list(category=...)
```
for all target categories.
```
skills_list(category='imported-codex')
```
should be empty or non-active.
```
skill_view(<key_skill>)
```
resolves to promoted category path.
Smoke-check critical runtime tools (e.g.
```
bird check --plain
```
).

Rollback support

Generate rollback script from manifest:
- Restore archive
```
SKILL.md
```
  from
```
SKILL.imported.md
```
  .
- Remove promoted target copies.

Store script under

~/.hermes/skill-promotions/rollback-*.sh

Pitfalls

Duplicate skill names across categories cause non-deterministic discoverability.
Moving without backups risks losing custom local edits in existing targets.
Forgetting to disable archive copies keeps the conflict alive.

Success criteria

User’s common skills appear in expected first-class categories.
Archive remains preserved and recoverable.
No duplicate active definitions for same skill name.
A documented rollback path exists.