Agentic Blog

Agentic Blog is a small, read-only command-line reader for public Naver Blog content. It is designed for focused retrieval: search public blogs and posts, inspect a blog and its category tree, list posts, read a post body and comments, browse topics, and view public buddies (neighbours).

It is anonymous by design. There is no account, login, API key, browser, cookie jar, profile, session, setup, status, or doctor command. The package uses plain HTTPS requests to Naver's public surfaces and does not attempt to access private or neighbour-only content.

Read DISCLAIMER.md before use. This is an unofficial tool. Public availability does not remove your responsibility to use the service and retrieved data lawfully and respectfully.

Install

Requires Python 3.11 or newer.

python -m pip install .
agentic-blog --version

Install a reviewed wheel with python -m pip install /path/to/agentic_blog-0.1.0-py3-none-any.whl. This release-ready worktree does not claim that version 0.1.0 has been published to PyPI.

The runtime dependency set is intentionally limited to httpx, platformdirs, and lxml. No browser dependency or optional browser extra is provided.

Quick start

Search public posts and save up to five results:

agentic-blog search "커피" --limit 5

The command writes UTF-8 JSON to the default data directory and prints a one-line save summary to stderr. To choose the destination explicitly:

agentic-blog search "커피" --limit 5 --output ./coffee.json

Use a search result's blog ID to inspect a blog, then list its posts:

agentic-blog blog example_blog --output ./blog.json
agentic-blog posts example_blog --limit 10 --output ./posts.json

Read a public post and its comments:

agentic-blog post "https://blog.naver.com/example_blog/123456789" --output ./post.json

post also accepts an m.blog.naver.com URL, a PostView.naver?blogId=...&logNo=... URL, or a two-part reference:

agentic-blog post example_blog 123456789 --no-comments --output ./post.json

See the installation guide, quick start, and CLI reference for fuller examples.

Commands

Run agentic-blog <command> --help for parser-level help. agentic-blog catalog emits a machine-readable JSON description of the installed CLI; agentic-blog schema prints its model fields, while agentic-blog schema --json emits JSON Schema draft 2020-12. Both meta commands are offline and write to stdout.

Read commands

All read commands accept --format {json,ndjson} (default json), --output PATH, --data-dir PATH, --no-redact, and -v/--verbose. Diagnostic redaction is enabled by default; --no-redact explicitly disables it for error messages only, never result files, and verbose diagnostics include the typed error class. --limit N applies to search, posts, buddies, and topic; it must be non-negative. --raw is accepted only by search, blog, posts, buddies, and topic. Neither post nor topics accepts --raw.

Important combinations are validated as usage errors:

• Search date bounds are server-side and available only with search --type post. --type id accepts neither --sort nor date bounds.
• posts --query cannot be combined with --category, --sort popular, or --notices; its text must not be empty.
• posts --notices cannot be combined with --sort popular, --category, or --query.
• A bare post number is not enough: provide a Naver post URL or both <blog_id> <log_no>.

Output and storage

Read commands write an array of schema-shaped objects to a file, never the retrieved content to stdout. Their stderr summary reports the item count, date range when available, stop reason, and saved path. JSON output is indented UTF-8 with ensure_ascii=False, preserving Korean text. NDJSON writes one UTF-8 JSON object per line.

Without --output, files are written beneath the platform user-data directory:

<platform user data>/agentic-blog/output/

Names include the command, a Unicode-safe identifier, UTC timestamp, and .json or .ndjson extension. Use --data-dir PATH to choose the application data directory for one command, or set AGENTIC_BLOG_DATA_DIR for the process. --output PATH takes precedence for the output file itself.

The output schema is generated from the installed models. Use:

agentic-blog schema --json

For schema and record guidance, see Output Schema.

Budgets and pacing

Every client enforces a non-bypassable 0.5-second minimum pause between requests. Each command run has a default budget of 100 requests, preventing accidental bulk collection. Pagination stops on the requested limit, a natural end, an empty search, or the request budget. search --type post sends --since and --until as server-side filters; they do not create a local date-boundary stop reason. The stderr summary identifies the stop reason.

There is intentionally no bulk crawl, daemon, batch mode, --profile, or --wait-on-limit flag. Compose small commands instead of attempting broad collection.

Chaining focused reads

Agentic Blog's commands are single-target primitives. A typical bounded investigation is:

1. search for a topic, blog name, or ID.
2. blog to inspect a selected public blog's profile and category tree.
3. posts or posts --query to narrow to relevant public posts.
4. post to read a selected body and comments.
5. buddies to follow only public neighbour links, or topics then topic to change discovery paths.

Save each result and pass only the identifiers needed to the next command. Do not treat chaining as authorization to collect a social graph or archive a service.

Limitations and exit codes

Only content that Naver makes readable to anonymous visitors is in scope. Private blogs, neighbour-only posts, deleted posts, suspended blogs, login-gated feeds, notifications, visitor statistics, and write actions are unavailable. The tool does not post, comment, like, add neighbours, or bypass access controls. Naver may change its public response or HTML structure; update the package rather than relying on a broken parser.

Exit code 2 is intentionally unassigned; it is not an authentication state.

Privacy and responsible use

Public content can still contain personal data. Keep outputs local, collect the minimum needed, avoid publishing or redistributing retrieved personal information, and delete outputs when no longer needed. Diagnostic raw output can expose more upstream fields; use it only for local troubleshooting and handle it with particular care.

Repository fixtures are synthetic and PII-free. Do not commit real captures, cookies, browser state, session data, credentials, or output files. Read SECURITY.md, DISCLAIMER.md, and the Security and Privacy guide.

Contributing and policies

Contributions are welcome under CONTRIBUTING.md. Please follow the Code of Conduct, report vulnerabilities through SECURITY.md, and review the FAQ and troubleshooting guide.

License

MIT