v1.0.0

The first release of conflu-c-cli, a command line client for Confluence Cloud for people at a terminal and for programs and AI agents. It was
built against Atlassian's published API descriptions. The test suite runs against a bundled mock Confluence Cloud server, and the read commands
were also checked, read-only, against a real Confluence Cloud site. conflu doctor is the check against your own site and token.

Added

• Read and navigate Confluence Cloud. conflu page, conflu pages, conflu tree, conflu versions, conflu search, conflu comments,
  conflu labels, conflu attachments, conflu spaces and conflu space read pages, trees, histories, search results, metadata and spaces.
  conflu page --body-only prints nothing but the body, to save it in a file. conflu download saves attachments and keeps credentials away from
  the media host that serves the file. conflu page and conflu tree also take page:KEY/Title (and page:Title with --space), the form in
  which the Markdown shows a link to another page; when the space KEY exists but holds no page of that name, the whole text is used as a title in
  the current space (page:CI/CD pipeline).
• Read a page in parts. conflu page --outline lists the headings with the size of each section, --section TITLE|N reads one, --max-chars N
  cuts at the end of a paragraph and says so (body.truncated, body.totalChars). On a real page of 17,800 characters the outline and the one
  section needed were 5 % of it. A code block ends only at a closing fence with at most three spaces in front of it (as in CommonMark), so a line
  inside a code block is never taken for a heading or a cut by --outline and --max-chars.
• Search that works for questions. conflu search turns the text into words and "phrases", leaves out filler words (English and German) and
  the punctuation at the end of a word, and says what it left out (terms, ignored). The words of a plain search form one query
  (text ~ "a b"), and a phrase is a clause of its own. When no page holds every word it searches again for the pages that hold some and says so
  (relaxed, match); --match all|any|auto chooses. When nothing is found, notes says what to try. A title is shown as written (Q&A).
• Hits that can be judged without opening the page. Each hit has its path (the parent pages), labels and version; --snippets N reads the
  first N pages with one request and shows the passage around the words (snippet, matched), cut from the page as conflu page shows it, so
  people stay out. Wildcards in the words of a search (a*b*c*) cannot slow the snippets down on a hostile page, and a word that occurs often
  cannot hide a rare one from the passage that is shown.
• Safe page changes. conflu update, conflu move, conflu comment, conflu label, conflu attach and conflu create change pages with
  --dry-run previews. conflu update supports --append, --prepend, --if-version and --if-unchanged-since; Markdown replacements are
  blocked unless --force when the page holds constructs Markdown cannot express.
• Markdown publishing and offline mirrors. conflu import-md and conflu import-md-dir publish Markdown files with front matter (title,
  space, parent, page_id, labels, message). conflu export writes a local mirror of a space or page tree with front matter, lossy
  markers and index.json, in Markdown or storage format; every file gets a path: line and every entry of index.json a path, which the
  import ignores. An unedited export imports again as a no-op, and a Markdown replacement of a body that holds what Markdown cannot say (also
  redacted-text: hidden e-mail addresses and personal links) is refused without --force. A title survives the round trip: export writes it
  exactly as it is, double spaces included, so that importing the file again does not rename the page; a title that had to be redacted is refused
  on import, and an e-mail address in a title never reaches a file name. conflu export and conflu import-md-dir are two one-way commands, not
  a sync: an import does not notice changes made in Confluence after the export. The help of both commands, the skill and the guides say so and
  give a safe routine.
• Program and agent support. --json on every data command, CONFLUENCE_OUTPUT=json for a whole session, fixed exit codes, JSON errors on
  stderr, conflu commands --json as a full machine-readable catalog, and the shipped skill skills/conflu-c-cli/SKILL.md. conflu help,
  conflu how-to, conflu changelog and conflu skill all work offline.
• Tool descriptions for agents. conflu commands --format mcp [--only read|write|all] prints the commands an agent may use in the shape of an
  MCP tools/list answer: name, description, JSON Schema for the input, annotations (read-only, destructive, idempotent) and _meta.conflu for
  the command line. Commands for the person (the token, the settings, the default space) and the raw api are never listed. Every command says
  when to use it (whenToUse in the catalog, "When to use" in the help).
• A guide to finding information, conflu how-to find-information, for people and agents: which command, how the search works, reading in
  parts, a mirror to search together with the code, a loop that spends little, and a checklist for when nothing is found.
• Connection setup and diagnostics. conflu setup-token, conflu remove-token, conflu use, conflu current, conflu clear-space,
  conflu config, conflu ping and conflu doctor configure the connection, save the token safely, remember a current space and explain setup
  problems step by step. A certificate that no known authority vouches for (unable to get local issuer certificate) is reported with the chain
  the server presented and what to do about it, and is not retried. A 401 or 403 from an address that is not an atlassian.net address (a custom
  domain of the site, where Atlassian does not read API tokens) names the gateway address https://api.atlassian.com/ex/confluence/CLOUD_ID as the
  way out, and conflu doctor warns about such an address before anything fails.
• Settings for real enterprise networks. CONFLUENCE_BASE_URL, CONFLUENCE_EMAIL, CONFLUENCE_API_TOKEN, CONFLUENCE_SPACE,
  CONFLUENCE_CA_BUNDLE (adds authorities to the system store), CONFLUENCE_TLS_STRICT, CONFLUENCE_INSECURE, CONFLUENCE_ALLOW_HTTP,
  CONFLUENCE_HIDE_USERS, CONFLUENCE_CONNECT_TIMEOUT, CONFLUENCE_MAX_TIME, CONFLUENCE_RETRY and CONFLUENCE_OUTPUT cover connection setup,
  privacy, TLS, timeouts and retries.
• Escape hatches and packaging. conflu convert previews the Markdown and storage-format converters, and conflu api exposes REST calls below
  the wiki root (api/ and rest/) with guarded writes (--write, --dry-run, no general DELETE). python install.py installs, updates and
  uninstalls the program, its docs, the skill and the examples.
• Offline documentation and tests. A README, architecture notes, how-to guides, generated reference documents, examples and a test suite
  around a mock Confluence Cloud server (python -m unittest discover -s tests -t ., python scripts/generate_docs.py,
  python scripts/check_release.py).

Security

• Private by default. People, account ids, e-mail addresses and personal links are hidden unless --with-users is allowed and given;
  CONFLUENCE_HIDE_USERS=true makes that a rule.
• A link to a profile counts as a mention. /wiki/people/ID and /wiki/spaces/~ID, absolute or as the page writes them, are shown as @user
  in the Markdown and the text of a page, in the snippets and in an export, and a Markdown link to one is removed with its text. Ordinary paths
  that merely contain a folder called people are left alone.
• Search excerpts name nobody by default. The search index writes the names of mentioned people into the excerpt as plain text, so
  conflu search leaves the excerpt out (null in the JSON) unless you add --with-users, which CONFLUENCE_HIDE_USERS=true refuses.
• The snippets of a search are cut from the page as conflu page shows it, so a mention is @user and an e-mail address [email]; the path of a
  hit is redacted like every title.
• Credentials stay with Confluence. The token is never accepted on the command line, is shown masked, and is taken out, together with the
  account e-mail, of what Confluence echoes in an error. Credentials go only to the configured address: a download follows the redirect to the
  media host without them, and conflu api refuses a path that names a host or contains ...
• Text written by others cannot drive the terminal. Control, escape and C1 characters are removed from text output, error messages and
  progress messages, and written as \u escapes in JSON.
• Files stay where you point them. conflu download and conflu export make file names safe (no folders, no Windows device names, no
  invisible or right-to-left override characters), refuse names that would end up as the same file, and need --force to overwrite.
• No script links. A Markdown link or image whose address uses a scheme other than http, https, mailto, ftp, ftps or tel (javascript: in any
  spelling, data:, file: and so on) loses the link and keeps its text.
• Bounded input. An answer of Confluence larger than 64 MiB once decoded, or a file larger than 512 MiB, is refused; compressed answers are
  decoded under the same cap; content nested too deeply is refused with a clear error.