saubakirov/kz-it-telegram-list/kz-add
Claude Code and agent skill by saubakirov/kz-it-telegram-list
- community
Checking that the SKILL.md file is still in the repository.
Preview and apply evidence-backed Telegram candidates with exact owner approval
Install
npx vskill@latest install saubakirov/kz-it-telegram-list/kz-addSKILL.md
KZ Add — Verified Candidate Intake
Authority
This is a CL operation. Run it only when the owner literally invokes /kz-add with arguments. Receive the complete literal argument string as $ARGUMENTS; never replace it with remembered or discovered candidates. Parsing, probing, and preview are non-mutating. Catalog mutation requires a current durable owner record approving the exact payload/action set. A plan, TS, ONB, prior batch, or agent-written owner name is not approval.
Read AGENTS.md, the active approved TFW task, and data/communities.json. Do not open unrelated candidate sources. Never infer identity, type, count, date, category, relevance, commerciality, or localized copy.
Inputs
Accept exactly one source mode from $ARGUMENTS:
--text <UTF-8 text>for one URL or pasted URL/Markdown/text content;--file <UTF-8 path>for one supplied file.
Only fixed HTTPS root peer URLs on t.me, telegram.me, or telegram.dog can become candidates. Every Telegram-like occurrence, including reserved actions, invites, message links, malformed or duplicate forms, remains in the occurrence ledger with an explicit disposition.
Ordered Operation
- Confirm explicit invocation, active task authority, a clean understood worktree, and a durable
task evidence location. Hash data/communities.json, README.md, index.md, ru/index.md, and kk/index.md before work.
- Run
python scripts/validate_schema.py. Parse with
python scripts/kz_intake.py parse --text <text> --output <ledger.json> or the exact --file form. Stop if UTF-8/source accounting fails. If every occurrence is non-candidate (including a reserved-action smoke sentinel), report the dispositions and stop without network or apply.
- Run the matching
python scripts/kz_intake.py probe ... --output <observations.json>form.
Retain the machine output. One fetch is reconciled through the unchanged groups/channels/bots classifier; any inconsistent identity/type tuple remains unresolved and contributes no fact.
- Check case-insensitive collisions across live entries, archive, canonical identity, aliases,
and this input. Record mechanical facts separately from evidence-cited IT relevance, Kazakhstan relevance, commerciality, category, and explicit EN/RU/KK copy review.
- Assign every candidate exactly one proposed action: add, reject, duplicate, or unresolved. Only
an observed, target-bound, collision-free, eligible, locale-complete row may be proposed to add. Build all five expected generated outputs in an isolated staged project; run schema validation and python scripts/generate_readme.py --check there.
- Create the closed proposal bundle, then run
python scripts/kz_intake.py preview --bundle <bundle.json> --root <repo> --stage-root <stage> --output <preview.json> --render <preview.md>. Present the digest, complete table, exact ADD set, non-add reasons, evidence paths, and current controlled-path hashes. Stop and request approval of that exact payload/action projection.
- After an explicit current owner response, retain it durably and create a separate
kz-intake-approval/v1 envelope binding the payload digest, action digest, exact add IDs, owner, evidence reference, and time. Any edit, re-observation, stale date, baseline change, or subset choice requires a successor preview and new approval.
- Recheck current authority, target freshness, collision/eligibility, staged schema/currency, and
controlled-path B/A/X state. Apply only with python scripts/kz_intake.py apply --preview <preview.json> --approval <approval.json> --root <repo> --stage-root <stage> --pending <pending.json> --receipt <receipt.json> --owner-ref <exact-ref>. All-before may write staged bytes; all-after is an exact no-op; only a matching durable pending marker permits mixed-state recovery. Unknown or unmarked mixed state stops.
- Re-run schema validation, generator currency, relevant tests, and exact probes for added rows.
Hash all five controlled paths again and retain the separate receipt and final dispositions.
Outputs
- Lossless occurrence and grouped-candidate ledgers; immutable observation evidence.
- Human preview plus canonical payload digest; separate owner envelope, pending marker, and receipt.
- One explicit disposition per occurrence/candidate and exact before/after controlled-path hashes.
- Only exact owner-approved rows that still pass every gate in source JSON and generated outputs.
Failure Conditions
- Any occurrence disappears, an invalid URL is salvaged, or bare handles are admitted as URLs.
- A foreign/ambiguous/private/dead/wrong-type target exposes an approved fact.
- Editorial positives, translations, counts, dates, names, or authority are guessed.
- Preview, owner action, staged bytes, controlled hashes, or applied bytes can diverge.
- A partial/unapproved set writes, schema/currency fails, or unrelated work is staged.
Hard Stop
Never mutate before exact current owner approval. Never archive, release, tag, push, or contact an external service as a side effect. On uncertainty, preserve evidence and dispositions, report the specific blocker, and stop.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingtasks/2026/20260828-201343__catalog_intake_commands/phase-b/evidence/successor-1/apply-stage/.agents/skills/kz-add/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/saubakirov/kz-it-telegram-list/kz-add)