Store evaluation

Store Evaluation

Score a proposed source store against five dimensions. Stores below the threshold are rejected. Stores above are added to sources/supported-stores.json via source-inventory.

When to Use

• A user invokes /add-source with a candidate store
• Periodic re-evaluation of borderline entries (every 6 months)

The Five Dimensions

Score 0–2 per dimension. Maximum score is 10. Threshold for inclusion: ≥ 7.

1. Maintenance Health (0–2)

Score	Criterion
0	Repo archived, no commits in 12+ months, last release > 2 years old
1	Active but low velocity — commits in last 6 months, no recent release
2	Active maintenance — commits in last 90 days OR active issue triage OR recent release

2. Adoption Signal (0–2)

Score	Criterion
0	Solo project, < 50 stars, no external contributors
1	Some adoption — 50–500 stars, occasional external contributors
2	Real adoption — > 500 stars OR multiple maintainers OR sustained external contribution

(Stars are noisy; weight them with maintenance + license signal.)

3. License Clarity (0–2)

Score	Criterion
0	No license, custom restrictive license, or unclear terms
1	Permissive license but with vendor-specific restrictions
2	Clean OSI-approved license (MIT, Apache 2.0, BSD, ISC, etc.)

4. Catalog Fit (0–2)

Score	Criterion
0	Off-topic — no clear use case for an AI agent that consumes plugins from this Mall
1	Tangential — useful in narrow scenarios
2	Direct fit — agents that use this Mall would install plugins from this source in normal workflows

5. Documentation Quality (0–2)

Score	Criterion
0	No README, broken docs, or installation steps unclear
1	Basic README, install instructions present, examples sparse
2	Quality README + examples + clear installation + troubleshooting

Decision

Total score	Verdict
9–10	Strong accept — add to the registry
7–8	Accept — add with a note of any caveats
5–6	Defer — request more signal (stars, contributors, time) and re-evaluate in 90 days
0–4	Reject — document the rationale in docs/curation-log.md

On accept: add the entry to sources/supported-stores.json per source-inventory. The Mall does NOT clone or modify the upstream repo — it scores it and surfaces it in the catalog; agents install directly from upstream at a pinned ref.

Output Template

# Store Evaluation — <store name>

**URL**: <repo URL>
**Date**: YYYY-MM-DD
**Evaluator**: <name>

## Scores

| Dimension | Score | Notes |
|---|---|---|
| Maintenance Health | x/2 | ... |
| Adoption Signal | x/2 | ... |
| License Clarity | x/2 | ... |
| Catalog Fit | x/2 | ... |
| Documentation Quality | x/2 | ... |
| **Total** | **x/10** | |

## Verdict

Strong accept | Accept | Defer | Reject

## Rationale

<2-3 sentences>

## Proposed registry entry (if accepted)

```jsonc
{
  "name": "<kebab-name>",
  "remote": "https://github.com/<org>/<repo>",
  "pluginDir": "<plugins | skills | . | other>",
  "quality": "<official | community-curated | community | domain | reference>",
  "provenance": false,
  "license": "<SPDX-id or null>",
  "added_at": "YYYY-MM-DD"
}

ACT Pass Trail

<markers> ```

Anti-Patterns

Anti-pattern	Correction
Scoring on stars alone	Stars are one signal of five; weight with maintenance and fit
"I like it" override	The scorecard is the contract. Document overrides as ADRs
Rejecting without recording	Every rejection sets precedent; document or future similar proposals will repeat the cycle

Falsifiability

This scorecard needs revision if any of the following occur within 90 days:

• Accepted stores (score >= 7) are repeatedly pruned for quality or maintenance failures
• Deferred stores (score 5-6) consistently pass later without new evidence, indicating scoring too strict
• Reviewers produce materially different scores for the same store with no documented rationale

Track these in docs/curation-log.md tagged [STORE-SCORING].

Related

• source-inventory — adding accepted stores to the registry
• staleness-discipline — pruning gates
• /add-source prompt