# Pjf > Professional Judgment Filter — a reusable decision module for deciding whether content from a private communication source (direct messages, private channels or group chats, email received or sent, meeting transcripts, or any other non-public exchange) may be used in a write to a more-visible destination: a task or ticket, a project tracker or board, a wiki or doc page, or a draft reply. Apply this skill whenever you process private-source content and must decide what to include. Use it in any scheduled task, capture bot, watch, digest, or follow-up drafter that reads private communications before writing to a shared or public destination. If you are about to write content sourced from a private message, email, private channel, or transcript into any destination more visible than the source — run the PJF first. - Skill: `ebellefontaine/pjf` (Agent Skill) - Install (CLI): `npx skillmds@latest add ebellefontaine/pjf` - Raw SKILL.md: https://api.skillmd.com/api/skills/ebellefontaine/pjf/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Productivity - Author: ebellefontaine (https://skillmd.com/u/ebellefontaine) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/ebellefontaine/pjf --- # Professional Judgment Filter (PJF) This module answers one question: **Should this content from a private source be used in a write to a more-visible destination?** Apply it to every piece of content that came from a private source — a direct message, a private channel or group chat, email (received or sent), a meeting transcript, or any other non-public exchange — before writing anything to a destination more visible than that source: a task or ticket, a project board, a wiki or doc, a draft message, or any shared workspace. The principle is source-to-destination visibility, not any particular tool. If the destination can be seen by more people than the source, the content is crossing a visibility boundary, and the PJF governs that crossing regardless of which apps are involved. --- ## The Three Gates Content passes only if **ALL THREE** conditions are true. Fail any one → skip the content entirely. ### Gate 1 — RELEVANCE The content must directly reference tracked work: a specific task, project, work-item or ticket key, sprint, integration, vendor, known dependency, or a named participant in active work. **Skip if:** off-topic conversation, general banter, tool complaints unrelated to a tracked item, personal or social messages, or anything that doesn't connect to a specific piece of active work. ### Gate 2 — PROFESSIONAL INTEGRITY If this content were extracted and written to a more-visible destination, it must accurately represent a work event and preserve the professional reputations of everyone named. **Skip if any of the following apply:** - Jokes, sarcasm, or hyperbole that could be misread as fact (e.g., "we should burn it all down and start over") - Venting, ranting, or informal opinions that don't represent a real decision - Anything that would embarrass a named colleague if seen out of context - Credentials, secrets, API keys, or sensitive security/access details - Private personal discussions unrelated to work delivery The test: *Would a reasonable colleague be comfortable seeing this written as a factual work note?* If no, skip it. ### Gate 3 — TASK VALUE The content must add at least **one** of these signals: - A decision made or a decision needed - A named point of contact or stakeholder - An implementation detail, integration requirement, or technical constraint - A deadline, milestone, goal, or metric that was set or changed - A blocker identified or resolved - An explicit commitment, ETA, or status change given by a named participant - An action item with a named owner **Skip if:** the message is purely informational background with no one committing to anything, making a decision, or surfacing a constraint. --- ## Output: What to do after filtering ### If content FAILS any gate Skip entirely. Do not: - Create a task or item from it - Reference or paraphrase it in a write - Use it as context in a draft response - Log it in a run summary with enough detail to expose the private content Log only: `[skipped — PJF: failed]` so the omission is traceable without revealing the content. ### If content PASSES all three gates Extract the **professional fact only**: - State the work-relevant finding in clean, neutral language - Strip informal language, emotional tone, jokes, and off-the-cuff phrasing - Never quote private-source text verbatim — paraphrase to the factual substance - Attribute by source type, not by quoting. Cite where it came from, not what was said: `(per direct message)`, `(per email)`, `(per meeting)` — never by reproducing the message **Examples:** | Raw private content | What to write | |---|---| | "ugh the stakeholder keeps being difficult about the analytics timeline, they keep pushing it to next quarter" | `Stakeholder has pushed the analytics integration target to next quarter. (per direct message)` | | "lol we're literally never going to finish this lmao" | Skip — Gate 2 fails (hyperbole/joke) | | "hey can you grab lunch tomorrow?" | Skip — Gate 1 fails (not work-relevant) | | "I'll have the reporting API spec ready by EOD Friday" | `Owner committed to reporting API spec delivery by EOD Friday. (per email)` | | "the password for the staging env is " | Skip — Gate 2 fails (credentials) | --- ## Applying it during a run When a task or workflow references the PJF, apply it per-item before any write, item creation, or draft composition. The filter is not a one-time pass at the start of a run — evaluate each candidate piece of content independently. If an entire private source fails Gate 1 (nothing in this thread touches tracked work), you can skip scanning the rest of that thread for that run. If a private source partially passes (some messages pass, some fail), extract only the passing content. Do not let a passing message pull in surrounding context that would fail. --- ## Edge cases **"This might be sarcasm"** — when in doubt, treat it as failing Gate 2. Don't make judgment calls that could expose a colleague to embarrassment. **Borderline TASK VALUE** — if you're unsure whether something adds value, ask: *Does this change what someone would do next?* If yes, it passes Gate 3. If it's just background color with no action implications, skip it. **Multiple signals in one message** — extract only the passing signals. A message can pass for one fact and fail for another in the same sentence. Extract the fact, leave the rest. **The operator's own outgoing messages** — apply the same filter to messages sent by the person on whose behalf you are working. Their own commitments made in email or messages are high-value signals (Gate 3 passes for any commitment), but their informal venting or off-the-cuff remarks still fail Gate 2.