verify-skrya
Skill8 starsChanged 27 days agoMITView on GitHub
Verify Skrya's real CLI from this checkout in isolated HOME and data roots, retaining command/output/artifact evidence. Covers digests, legacy numeric analysis, Threads, retrieval/ingest v2, template validation, native cards and frozen references, delivery preconditions, source health/cache/ingest-prune maintenance, ingest v1 migration, and install/uninstall. Use after changes to CLI behavior, templates, fixtures, delivery or installation.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- deep-analysisSkill
- digestSkill
- request-curationSkill
- skryaSkill
- source-curationSkill
- topic-curationSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 24
No build step. Prereqs: python3 ≥ 3.10 with PyYAML importable (python3 -c "import yaml").
- Distinctive: line 37
Run first, and again whenever anything looks off.
- Very specific: line 91
Check side effects across the entire data root, not just runs/: compare before/after manifests of every file hash, directory entry, and …
- Very specific: line 94
Before delivery-precondition tests, remove inherited SKRYA_FEISHU_APP_ID, SKRYA_FEISHU_APP_SECRET, and SKRYA_FEISHU_CONFIG from the …
- Very specific: line 105
Always run cleanup in a finally/shell trap, including after failed drives.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
verify-skrya
Skrya is a skill pack: agents read SKILL.md files and call one deterministic surface, the Python CLI python3 -m skrya_orchestrator.main <command> (package src/skrya_orchestrator/). That CLI is what this skill drives. There is no server, port, or UI.
Conversational configuration flows and article drafting are agent workflows, not standalone CLI commands. Native cards can be rendered locally; actual Feishu delivery and desktop/mobile appearance require an explicitly approved sandbox application and recipient.
The helper scripts/verify-skrya (executable, bash) wraps every call so it:
- runs from the repo root with
PYTHONPATH=<repo>/src→ code under test is this checkout, never the pipxskryaon PATH (~/.local/bin/skryais a separate, possibly stale install); - sets
SKRYA_DATA_ROOT=/tmp/verify-skrya-<run>/data→ never touches~/.skrya; - sets
HOME=/tmp/verify-skrya-<run>/home→ install/uninstall/data-root --set --scope homewrite there, not into~/.codex,~/.claude,~/.skrya/config.json; - passes
PYTHONUSERBASEthrough so PyYAML from the user site still imports under the fake HOME; - records
cmd.txt,stdout.txt,stderr.txt,exit_codeper call.
Use a fresh isolated run for each feature recipe; drive its short-lived CLI processes serially. Run doctor before the first drive in every run and after every nonzero exit, including expected error cases. If doctor fails, do not continue driving until the cause is understood and health restored.
Launch
No build step. Prereqs: python3 ≥ 3.10 with PyYAML importable (python3 -c "import yaml").
V=.agents/skills/verify-skrya/scripts/verify-skrya # from repo root
eval "$($V launch)" # exports VERIFY_SKRYA_RUN=<timestamp>-<pid>launch creates /tmp/verify-skrya-$VERIFY_SKRYA_RUN/{data,home,work}, an owner marker, and seeds data/topics/ with a copy of tests/fixtures/topics/ (agi-news, k-entertainment, new-energy-vehicles). Ready = stderr line verify-skrya: ready run=... scratch=... evidence=.... It also writes run.txt (git HEAD, dirty file count, seeded topics) into the evidence dir.
Each shell needs VERIFY_SKRYA_RUN exported; if lost, recover it with ls tmp/verify-skrya/evidence/.
Doctor
Run first, and again whenever anything looks off. Read-only.
$V doctorRequires all [OK]: scratch dir owned by this run, evidence dir exists, skrya_orchestrator imports from <repo>/src (not site-packages/pipx), version prints Skrya version: + local revision, data-root reports Source: SKRYA_DATA_ROOT pointing inside the scratch dir. Any [FAIL] → exit 1; fix before driving. A module/yaml import failure means PyYAML is missing for this python3 (override interpreter with VERIFY_SKRYA_PYTHON=/path/to/python).
Drive
$V skrya <label> -- <cli args...> # prints stdout, stderr tail, exit code; returns CLI exit code
$V snapshot <label> # copies data/runs/; lists data/home and top-level work symlinks
$V path data|home|work|scratch|evidencePass --root . for ordinary commands. For installation, workspace data-root config and uninstall, first create an archived checkout under the run's work/ and use it as --root (see features/data-root-install.md); never let those commands edit the real checkout's packaging or instructions. Do not pass --data-root except to verify precedence. Put inputs under $($V path work).
Core commands (exact flags from main.py):
| User action | Command |
| Daily digest (writes artifacts) | digest --root . --topic <id-or-name> [--sample] [--force-refresh] |
| Test run / preview (no artifacts) | digest --root . --topic <id> --preview (alias --test-run) |
| Legacy numeric analysis | deep-analysis --root . --topic <id> --event-number N; current agent follow-ups use frozen resolution below |
| Thread timeline | thread --root . --topic <id> --thread <id-or-name> |
| Rebuild threads | refresh-threads --root . --topic <id> |
| Retrieval request | retrieval-request --root . --topic <id> [--since ISO] [--until ISO] |
| Record normalized retrieval | ingest --root . --topic <id> --file <skrya.ingest.v2 json> [--raw-file f] |
| Source health report | source-health --root . --topic <id> [--refresh]; --refresh runs one live collection for the explicit topic (network) |
| Cache maintenance | cache --root . --clear-http-cache / --clear-translation-cache [--translation-language L] / --cleanup-expired; no topic required; with no flag it prints usage and exits 2 |
| Prune ingest batches | ingest-prune --root . --topic <id> [--keep N] [--apply]; dry-run by default, --apply deletes only selected archived batch files |
| Migrate ingest state to v2 | upgrade --root . --migrate-ingest-v2 [--topic <id>]; rewrites only current latest-ingest files and keeps latest-ingest.v1-backup.json next to them |
| Template save gate | validate-digest-template --root . --file <md> |
| Data root | data-root --root . [--set P --scope home|workspace] [--migrate] |
| Install / uninstall | install-skill-pack --root "$R" --host codex|claude|openclaw ... / uninstall-skill-pack --root "$R" ...; R is the archived scratch checkout |
| Native card bundle | render-card --root . --file <digest.md> --format feishu-v2 -o <bundle.json> |
| Frozen local reference | resolve-reference --root . --topic <id> --artifact <digest.md> --reference N-xx|T-xx |
| Delivery preflight / send | delivery-check / deliver --channel feishu --topic <id> --target <bound-target> --binding <id>; deliver also requires --file |
| Received reference | resolve-reference --root . --topic <id> --binding <id> [--message-id <id>] --reference N-xx|T-xx |
Topic arguments accept the id or display name. Legacy analysis/thread errors may end in a Python traceback; digest/render/delivery/reference commands instead print a concise failure and exit 1. Inspect both exit code and captured stdout/stderr, not one assumed error format.
Feature recipes with exact commands and expected observations: features/README.md. Open features/source-maintenance.md for isolated RSSHub, health, cache, lifecycle maintenance, and full-data-root preview equality.
Evidence
Location: <repo>/tmp/verify-skrya/evidence/<run>/ (tmp/ is gitignored). One numbered step dir per helper call: NN-<label>/{cmd.txt,stdout.txt,stderr.txt,exit_code}; snapshots: NN-snapshot-<label>/{tree.txt,runs/}.
Proof standards:
- Drive the real CLI entry point (
python -m skrya_orchestrator.main), notIntelligenceServicemethods or test helpers. - Capture the action and the resulting state: the call's stdout/exit code and a
snapshotof what it wrote (runs/<topic>/digest-*.md,latest-digest.mdsymlink target,latest-digest-events.json,deep-analysis-N.md,threads/latest-threads.json,ingest/...). - For mutations, prove with a second read: e.g. after
digest,deep-analysisresolving the same item number; afteringest, adigest --previewwhose items and扫描时间范围come from the ingested file. - Check side effects across the entire data root, not just
runs/: compare before/after manifests of every file hash, directory entry, and symlink target. Usefeatures/source-maintenance.mdfor the runnable recipe; it covers topic health, event state, HTTP and translation caches as well as digest artifacts. - Network boundaries: live RSS retrieval and translation of non-Chinese text can use the network. Digest no longer adds a version-check footer.
doctorinvokesversion, whose upstream check may use the network. Say when fixtures were used. - Plain Markdown ends in five actions without model/version/Agent/LLM execution metadata. Native card date badges and layout are checked separately from archive dates.
- Before delivery-precondition tests, remove inherited
SKRYA_FEISHU_APP_ID,SKRYA_FEISHU_APP_SECRET, andSKRYA_FEISHU_CONFIGfrom the helper's environment. Never borrow real credentials or bindings just to make a path reachable.delivery-checkcan contact the API even though it sends no message. - With explicit sandbox authorization, a temporary local lark-cli transport boundary may drive the real CLI entry point while the CLI owns authentication. Follow
features/feishu-delivery.md, capture the exact runner, and report separately what orchestration passed and which default-auth/transport paths were not exercised. A standalone CLI card send is not sender proof.
Cleanup
$V cleanup # or: $V cleanup <run-id>Removes only /tmp/verify-skrya-<run>/ and only if its owner marker matches the run id; refuses otherwise. It also deletes <repo>/.skrya/config.json when (and only when) that file points into this run's scratch dir, since data-root --set ... --scope workspace writes it in the real checkout. No processes are started, so nothing to kill. Evidence under tmp/verify-skrya/evidence/<run>/ is kept; confirm with ls tmp/verify-skrya/evidence/<run>/. Run cleanup after failed iterations too; orphaned runs show up as ls -d /tmp/verify-skrya-*, and you clean each one with $V cleanup <run-id>. Delete old evidence by hand only when the user no longer needs it.
Always run cleanup in a finally/shell trap, including after failed drives. Before cleanup, enumerate the evidence files and record their hashes; afterward confirm they still exist unchanged at the named evidence directory. Scratch removal is not evidence retention proof.
- Build/install operations regenerate host assets under the supplied root. Use the archived root so no generated assets or product instructions are changed in the real checkout.
Gotchas
- Never call bare
skryaor an unisolated Python entry point for verification: real data and host skill directories would be used. The approved temporary Feishu CLI-boundary runner is the sole exception to the helper invocation shape, not to isolated HOME/data-root, doctor, evidence or cleanup requirements. deep-analysisandthreadread artifacts from the last non-previewdigest; on a fresh run they fail withDigest artifact not found/Thread artifact not found. Run a realdigestfirst.uninstall-skill-pack --mode data-keep-skills|completedeletes the resolved data root — under the helper that is the scratchdata/, so later steps lose the seeded topics. Do it last or re-launch.- Complete uninstall may strip routing-note blocks in the supplied root. Use only the archived scratch checkout for this path; do not rely on the real repository currently lacking a marker.
install-skill-pack --host autopicks hosts by marker dirs in HOME; the scratch HOME has none, so it falls back tocodex. Pass--hostexplicitly.
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.

