curator-evolution

Name: curator-evolution
Author: pingchesu

27stars3forks

How to install curator-evolution?

pingchesu/hermes-curator-evolver/curator-evolution

$npx -y skills add pingchesu/hermes-curator-evolver --skill curator-evolution

Installs into the current project.

Summary

Use when interpreting Hermes Curator Evolver evidence reports, historical session backfill, proposals, verifier results, candidate search, auto-run output, timer installs, or guarded apply manifests.

Files · 1

View on GitHub

SKILL.md

1	# Curator Evolution
2
3	Hermes Curator Evolver starts from evidence and keeps mutation guarded. Reports and proposals are review artifacts. `auto-run` can make skills actually improve, but only through low-risk bounded managed blocks, reference spillover for bulky evidence, and guarded apply.
4
5	## Interpretation Checklist
6
7	1. Separate evidence from conclusions.
8	2. Repeated tool errors suggest a possible missing pitfall or verification step in a skill.
9	3. Repeated skill reads suggest the skill is active and worth keeping discoverable.
10	4. A single failure is not enough evidence to rewrite a skill.
11	5. If the skill was correct but the agent ignored it, improve triggers/descriptions only when evidence repeats.
12	6. Candidate search is advisory; embedding/reranker models only find candidates and do not decide edits.
13	7. `backfill-sessions` can import existing Hermes `session_*.json` transcripts into evidence; it is model-free and does not mutate skills.
14	8. `auto-run --semantic-candidates` and `--rerank-candidates` are explicit opt-ins that only reorder evidence-eligible candidates.
15	9. Guarded apply requires approval, backup, verifier/validation pass, and rollback.
16	10. `auto-run` mutates only when both `--apply-low-risk` and `--approve-auto-apply` are set.
17	11. Even with write flags, unattended auto-apply writes only local agent-created skills. Official/bundled skills (`.bundled_manifest`), hub-installed skills (`.hub/lock.json`), plugin-provided skills, `skills.external_dirs`, pinned skills, and unknown sources are skipped.
18	12. Core-name allowlists only operate inside that local agent-created source boundary; they do not override provenance.
19	13. Size guardrails target a 90k `SKILL.md` soft cap, spill bulky autorun evidence into `references/`, and skip unattended writes when the target skill is already over the 100k hard cap.
20	14. `install-auto --enable` creates a native user scheduler: systemd user timer on Linux, launchd LaunchAgent on macOS; remove it with `uninstall-auto` before plugin uninstall.
21
22	## Safe Next Actions
23
24	- Run `hermes-curator-evolver bootstrap` for the default one-command setup: backfill recent sessions and enable the daily safe autorun scheduler.
25	- Run `hermes-curator-evolver bootstrap --semantic` only when the user explicitly wants model-assisted scheduler candidate ordering.
26	- Run `hermes-curator-evolver report --days 7` for evidence.
27	- Run `hermes-curator-evolver backfill-sessions --sessions-dir ~/.hermes/sessions --days 30 --format json` once when existing Hermes sessions should seed evidence.
28	- Run `hermes-curator-evolver propose --skill <name> --format json` for a dry-run proposal.
29	- Run `hermes-curator-evolver verify --proposal-file <proposal.json>` before considering apply.
30	- Use `hermes-curator-evolver candidates --query <text> --skills-dir <dir>` for dependency-free lexical candidate search.
31	- Use `hermes-curator-evolver auto-run --skills-dir ~/.hermes/skills --format json` to preview automatic improvements.
32	- Use `hermes-curator-evolver auto-run --skills-dir ~/.hermes/skills --semantic-candidates --rerank-candidates --format json` to preview model-assisted candidate ordering.
33	- Use `hermes-curator-evolver auto-run --skills-dir ~/.hermes/skills --apply-low-risk --approve-auto-apply` for actual low-risk bounded improvement; only local agent-created skills are writable, bulky evidence can spill to `references/`, and already-over-hard-cap skills are skipped.
34	- Use `hermes-curator-evolver install-auto --schedule daily --enable` for plug-in daily automation without Hermes core changes; portable schedules are `hourly`, `daily`, and `weekly`.
35	- Use `hermes-curator-evolver install-auto --schedule daily --enable --semantic-candidates --rerank-candidates` only when the user explicitly wants model-assisted scheduler candidate ordering.
36	- Use `hermes-curator-evolver uninstall-auto` to remove the optional scheduler.
37	- Use guarded apply only with an exact SHA256 and a reviewed content file.

People who liked this also viewed

6 picks

Type

curator-evolution

27stars3forks

View on GitHub

How to install curator-evolution?

pingchesu/hermes-curator-evolver/curator-evolution

$npx -y skills add pingchesu/hermes-curator-evolver --skill curator-evolution

Installs into the current project.

Summary

Use when interpreting Hermes Curator Evolver evidence reports, historical session backfill, proposals, verifier results, candidate search, auto-run output, timer installs, or guarded apply manifests.

Files · 1

View on GitHub

SKILL.md

1	# Curator Evolution
2
3	Hermes Curator Evolver starts from evidence and keeps mutation guarded. Reports and proposals are review artifacts. `auto-run` can make skills actually improve, but only through low-risk bounded managed blocks, reference spillover for bulky evidence, and guarded apply.
4
5	## Interpretation Checklist
6
7	1. Separate evidence from conclusions.
8	2. Repeated tool errors suggest a possible missing pitfall or verification step in a skill.
9	3. Repeated skill reads suggest the skill is active and worth keeping discoverable.
10	4. A single failure is not enough evidence to rewrite a skill.
11	5. If the skill was correct but the agent ignored it, improve triggers/descriptions only when evidence repeats.
12	6. Candidate search is advisory; embedding/reranker models only find candidates and do not decide edits.
13	7. `backfill-sessions` can import existing Hermes `session_*.json` transcripts into evidence; it is model-free and does not mutate skills.
14	8. `auto-run --semantic-candidates` and `--rerank-candidates` are explicit opt-ins that only reorder evidence-eligible candidates.
15	9. Guarded apply requires approval, backup, verifier/validation pass, and rollback.
16	10. `auto-run` mutates only when both `--apply-low-risk` and `--approve-auto-apply` are set.
17	11. Even with write flags, unattended auto-apply writes only local agent-created skills. Official/bundled skills (`.bundled_manifest`), hub-installed skills (`.hub/lock.json`), plugin-provided skills, `skills.external_dirs`, pinned skills, and unknown sources are skipped.
18	12. Core-name allowlists only operate inside that local agent-created source boundary; they do not override provenance.
19	13. Size guardrails target a 90k `SKILL.md` soft cap, spill bulky autorun evidence into `references/`, and skip unattended writes when the target skill is already over the 100k hard cap.
20	14. `install-auto --enable` creates a native user scheduler: systemd user timer on Linux, launchd LaunchAgent on macOS; remove it with `uninstall-auto` before plugin uninstall.
21
22	## Safe Next Actions
23
24	- Run `hermes-curator-evolver bootstrap` for the default one-command setup: backfill recent sessions and enable the daily safe autorun scheduler.
25	- Run `hermes-curator-evolver bootstrap --semantic` only when the user explicitly wants model-assisted scheduler candidate ordering.
26	- Run `hermes-curator-evolver report --days 7` for evidence.
27	- Run `hermes-curator-evolver backfill-sessions --sessions-dir ~/.hermes/sessions --days 30 --format json` once when existing Hermes sessions should seed evidence.
28	- Run `hermes-curator-evolver propose --skill <name> --format json` for a dry-run proposal.
29	- Run `hermes-curator-evolver verify --proposal-file <proposal.json>` before considering apply.
30	- Use `hermes-curator-evolver candidates --query <text> --skills-dir <dir>` for dependency-free lexical candidate search.
31	- Use `hermes-curator-evolver auto-run --skills-dir ~/.hermes/skills --format json` to preview automatic improvements.
32	- Use `hermes-curator-evolver auto-run --skills-dir ~/.hermes/skills --semantic-candidates --rerank-candidates --format json` to preview model-assisted candidate ordering.
33	- Use `hermes-curator-evolver auto-run --skills-dir ~/.hermes/skills --apply-low-risk --approve-auto-apply` for actual low-risk bounded improvement; only local agent-created skills are writable, bulky evidence can spill to `references/`, and already-over-hard-cap skills are skipped.
34	- Use `hermes-curator-evolver install-auto --schedule daily --enable` for plug-in daily automation without Hermes core changes; portable schedules are `hourly`, `daily`, and `weekly`.
35	- Use `hermes-curator-evolver install-auto --schedule daily --enable --semantic-candidates --rerank-candidates` only when the user explicitly wants model-assisted scheduler candidate ordering.
36	- Use `hermes-curator-evolver uninstall-auto` to remove the optional scheduler.
37	- Use guarded apply only with an exact SHA256 and a reviewed content file.

People who liked this also viewed

6 picks

Type