Skip to main content

agentic-bees

Use an AI agent to plan and perform bounded, on-demand Btrfs duplicate scans and authorized deduplication with agentic-bees; no resident daemon or external AI API required.

Informations de source

Dépôt
LIghtJUNction/agentic-bees
Dernière activité de la source
2 novembre 2026 à 10:59
Langue détectée de SKILL.md
anglais
Étoiles
0
Forks
0

Plan bounded Btrfs dedup with agentic-bees

Plans bounded Btrfs duplicate scans in an authorized directory, then applies authorized deduplication, without a resident daemon or external model API.

Uses

Use when you want an agent to plan and perform bounded, on-demand Btrfs duplicate discovery, and to apply deduplication only within a directory you have authorized. No resident daemon or external model API is required for this workflow. This skill does not present the project as a completed replacement for upstream bees.

Prerequisites

The supported target is Linux x86_64 with Btrfs and kernel openat2 (Linux 5.6 or newer). Use an already installed agentic-bees binary, or the repository's target/release/agentic-bees after a local build with Rust 1.88 or newer. Check --version and --help first; an installed v0.1.0 lacks newer development features. Do not install software or services unless requested. An external model API is not required. Documented commands were not executed here.

How to use

Determine the narrowest existing absolute directory the user authorized; do not substitute slash-root, the entire home directory, or a broader mount merely because it is accessible. Keep plans, output, and state outside the scanned tree, then run inspect and confirm the root is Btrfs, using current user permissions and never prepending unconditional sudo. Start with one bounded scan, review warnings, completeness, and candidates, and resume with the full saved plan when continuing. Discovery, an agent recommendation, write permission, or a model's apply_plan response does not authorize mutation; present the concrete plan and obtain an explicit apply decision unless that authorization was already given for a clearly defined finite scope. Documented commands were not executed here.

Limitations

The original post and the pinned skill both say the project is still in development and has not reached full upstream bees feature parity; residual-reference rewriting and allocation reclamation remain unfinished, and the README says this development tree must not be published as a completed bees replacement. An installed v0.1.0 lacks newer development features such as the resumable foreground service and optional extent discovery. Successful traversal does not establish that all duplicates were found. Apply validates identity, bounds, and content before kernel deduplication, can succeed only in part, and has no transactional rollback; bytes_deduped is kernel-processed logical bytes, never physical disk space reclaimed. Default run and scan operation only discovers; do not add --apply or --continuous to a finite discovery task. The optional decide command is CLI-only, may call a billable provider, is always advisory_only, and is not an MCP tool. The service, when used, runs in the foreground, defaults to one batch, and installs no startup unit. GPL-3.0-or-later, as stated in COPYING, requires corresponding source when you convey the program and gives no warranty. GitHub login LIghtJUNction differs from the posting handle LIghtJUNction_x; the public mapping and exact-path participation are recorded, and relationApproved remains false. Official same-version reuse is unknown without a production query. Plans contain paths, hashes, and metadata and must be retained as local private data outside the scanned tree. Do not send data to an external model API unless explicitly requested. If decide is used, send only allowlisted aggregate counters and completion status, never paths, hashes, warnings, or contents, and do not print or copy JEV_API_KEY into command arguments, logs, or configuration. Remove task-only plans, locks, and temporary indices after completion once they are no longer needed for recovery or audit. Do not substitute slash-root, the entire home directory, or a broader mount merely because it is accessible, and never prepend unconditional sudo or treat permission errors as authority to escalate. Do not edit host MCP configuration to expand access or enable mutation without authorization; client roots cannot extend the host's existing absolute roots, and without host --enable-apply, apply is listed but denied. Do not widen a narrower authorized directory to satisfy --extent-scan; that option requires an explicitly authorized complete subvolume root and kernel tree-search privilege, and returned filesystem addresses are not physical device offsets.

Options d'installation

Le prompt qui vérifie d'abord la source est sélectionné par défaut. Vous pouvez passer à une commande directe ou télécharger une copie locale.

Vérifiez les fichiers source

Lisez SKILL.md et les fichiers associés affichés par SkillsMP avant de décider de l'installer.

Affichage de SKILL.md

SKILL.md
Instructions source · Aperçu en lecture seule
nameagentic-bees
descriptionUse an AI agent to plan and perform bounded, on-demand Btrfs duplicate scans and authorized deduplication with agentic-bees; no resident daemon or external AI API required.