genome-analyzer
Skill2 starsChanged 4 months agoMITView on GitHub
Cluster genomes into families, retrieve by similarity+utility, run replay, propose patches
In this repository24 other files it gives its agents.
- AGENTS.mdAGENTS.md
- Academic Paper SummarizationSkill
- Accessibility WCAG AuditSkill
- API Error TroubleshootingSkill
- CI/CD Pipeline TroubleshootingSkill
- Competitor Analysis SynthesisSkill
- Customer Onboarding AssistanceSkill
- Dependency Audit ResolutionSkill
- Docker Image OptimizationSkill
- GDPR Data MappingSkill
- incident-distillerSkill
- Kubernetes Pod DebuggingSkill
- nve-genome-explorerSkill
- OWASP Input ValidationSkill
- repo-auditorSkill
- research-packagerSkill
- rule-patcherSkill
- Secret Leak RemediationSkill
- skill-curatorSkill
- skill-package-builderSkill
- SOC2 Policy VerificationSkill
- tama-community-summarizerSkill
- Technology Stack EvaluationSkill
- Ticket Triage and RoutingSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
After creating a new genome → cluster it into a family
- Distinctive: line 13
After fixing a bug → retrieve similar genomes to check for patterns
- Distinctive: line 14
Before promoting a patch → run replay on family members
- Very specific: line 23
Read FAMILY_INDEX.json for existing families
- Distinctive: line 33
If best match_score ≥ 5 → assign to that family
- Distinctive: line 34
If best match_score < 5 → create new family
- Very specific: line 93
Run cli/tama-genome-report.js to generate:
- Distinctive: line 134
Replay poverty: Not enough family members → promote with caveat "replay: not_run"
≈ “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.
Files
Genome Analyzer
Family-level analysis, retrieval, replay, and patch proposal for Failure Genomes.
When to Use
- After creating a new genome → cluster it into a family
- After fixing a bug → retrieve similar genomes to check for patterns
- Before promoting a patch → run replay on family members
- Periodically → generate health report
Capabilities
1. Family Clustering
Input: A new genome file (FG-NNNNNN.json) Process:
- Read
FAMILY_INDEX.jsonfor existing families - For each family, compute match score:
match_score = 0 + 3 if same failure_family + 2 if same violated_invariant + 1 per overlapping stack_tag + 1 per overlapping surface_tag + 2 if repair_operator in same neighborhood - If best match_score ≥ 5 → assign to that family
- If best match_score < 5 → create new family
Output: Updated FAMILY_INDEX.json
2. Similarity + Utility Retrieval
Input: Current incident context (stack tags, surface, failure type) Process:
- Filter genomes by overlapping context fingerprint
- Score each by:
relevance_score = similarity * 0.4 + utility_score * 0.6 - Sort descending
- Return top 5
Why utility matters: Semantic similarity alone returns "looks similar" genomes. Utility-weighted retrieval returns "similar AND useful" genomes. A genome with utility 0.9 that matches 3 tags beats a genome with utility 0.2 that matches 5 tags.
3. Replay Before Promotion
Input: Proposed patch + genome + family members Process:
FOR EACH family member genome:
1. Read the incident that caused it
2. Mentally simulate: would the proposed patch have caught this?
3. Score: PASS (would have prevented) or FAIL (would not have helped)
FOR 1-2 holdout genomes (different family):
1. Check: does the patch obviously break this scenario?
2. Score: SAFE (no regression) or REGRESS (patch causes harm)
COMPUTE:
family_pass_rate = pass_count / family_total
holdout_safe = all holdouts are SAFE
DECISION:
IF family_pass_rate >= 0.6 AND holdout_safe → PROMOTE
IF family_pass_rate >= 0.6 AND NOT holdout_safe → REVIEW (manual)
IF family_pass_rate < 0.6 → REJECT or REFINEOutput: Updated genome replay field + promotion_decision
4. Patch Type Recommendation
Based on the failure genome, recommend which patch type is most effective:
| Family Pattern | Recommended Patch |
| Invariant violation (rule was clear) | verifier_patch — add check |
| Missing knowledge (no rule existed) | rule_patch — add rule |
| Process gap (steps were skipped) | workflow_patch — add step |
| Capability gap (can't do X) | skill_patch — add capability |
| Stale documentation | doc_patch — update docs |
5. Health Report Generation
Run cli/tama-genome-report.js to generate:
{
"report_date": "2026-03-18",
"total_genomes": 7,
"total_families": 7,
"families_by_count": [
{ "family": "verification-skipped-before-done", "count": 1, "avg_utility": 0.9 }
],
"repair_operators_by_reuse": [
{ "operator": "add-production-verification-step", "reuse_count": 5 }
],
"patch_types_by_win_rate": [
{ "type": "rule_patch", "promoted": 6, "rejected": 0, "win_rate": 1.0 }
],
"stale_genomes": [],
"low_utility_genomes": [],
"transferred_genomes": {
"successful": 0,
"failed": 0
}
}Utility Decay Formula
After 30 days without use: score *= 0.9
After 90 days without use: score *= 0.7
After each reuse: score *= 1.1 (cap 1.0)
After prevention: score *= 1.2 (cap 1.0)
After negative transfer: score *= 0.5Anti-Patterns for Genome Analysis
- Over-clustering: Different failures merged into one family → keep families narrow
- Under-clustering: One family fragmented into tiny buckets → merge if same invariant
- Utility lock-in: Old genomes dominate → apply decay
- Negative transfer: Imported genome harms local repo → track and penalize
- Replay poverty: Not enough family members → promote with caveat "replay: not_run"
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.

