@forwardimpact/outpost 3.3.3 → 3.4.0

package/templates/.claude/skills/req-forget/references/locations.md

@@ -6,21 +6,21 @@ inventory.
 ## Knowledge base — direct notes
 
 ```bash
-ls -d "knowledge/Candidates/{Name}/" 2>/dev/null
-ls "knowledge/People/{Name}.md" 2>/dev/null
+ls -d "Knowledge/Candidates/{Name}/" 2>/dev/null
+ls "Knowledge/People/{Name}.md" 2>/dev/null
 
 # Common name variations
-ls "knowledge/People/{First} {Last}.md" 2>/dev/null
-ls "knowledge/People/{Last}, {First}.md" 2>/dev/null
+ls "Knowledge/People/{First} {Last}.md" 2>/dev/null
+ls "Knowledge/People/{Last}, {First}.md" 2>/dev/null
 ```
 
 ## Knowledge base — backlinks and mentions
 
 ```bash
-rg -l "{Name}" knowledge/
-rg -l "{First name} {Last name}" knowledge/
-rg -l "\[\[.*{Name}.*\]\]" knowledge/
-rg -l "{email}" knowledge/
+rg -l "{Name}" Knowledge/
+rg -l "{First name} {Last name}" Knowledge/
+rg -l "\[\[.*{Name}.*\]\]" Knowledge/
+rg -l "{email}" Knowledge/
 ```
 
 ## Cached email threads
@@ -47,13 +47,13 @@ rg -l "{Name}" ~/.cache/fit/outpost/state/ 2>/dev/null
 ## Drafts
 
 ```bash
-rg -l "{Name}" drafts/ 2>/dev/null
+rg -l "{Name}" Drafts/ 2>/dev/null
 ```
 
 ## Final verification
 
 ```bash
-rg "{Name}" knowledge/ ~/.cache/fit/outpost/ drafts/
+rg "{Name}" Knowledge/ ~/.cache/fit/outpost/ Drafts/
 ```
 
 Expected: only the erasure report matches.

package/templates/.claude/skills/req-forget/references/report-template.md

@@ -1,7 +1,7 @@
 # Erasure Report Template
 
 Audit trail for `req-forget` Step 4. Save to
-`knowledge/Erasure/{Name}--{YYYY-MM-DD}.md`.
+`Knowledge/Erasure/{Name}--{YYYY-MM-DD}.md`.
 
 **The report itself must not contain personal data** beyond the subject's name
 and the actions taken. Do not copy CV content, skills, or assessments into the
@@ -22,15 +22,15 @@ report — record only what was deleted, not what it contained.
 ## Actions Taken
 
 ### Deleted Files
-- `knowledge/Candidates/{Name}/brief.md`
-- `knowledge/Candidates/{Name}/CV.pdf`
-- `knowledge/Candidates/{Name}/screening.md`
-- `knowledge/People/{Name}.md`
+- `Knowledge/Candidates/{Name}/brief.md`
+- `Knowledge/Candidates/{Name}/CV.pdf`
+- `Knowledge/Candidates/{Name}/screening.md`
+- `Knowledge/People/{Name}.md`
 - {list all deleted files}
 
 ### Redacted References
-- `knowledge/Organizations/{Agency}.md` — removed backlink
-- `knowledge/Candidates/Insights.md` — removed {N} bullet(s)
+- `Knowledge/Organizations/{Agency}.md` — removed backlink
+- `Knowledge/Candidates/Insights.md` — removed {N} bullet(s)
 - {list all redacted files and what was removed}
 
 ### Cached Data Removed
@@ -57,7 +57,7 @@ The following are outside this tool's reach:
 ## Verification
 
 ```bash
-rg "{Name}" knowledge/ ~/.cache/fit/outpost/
+rg "{Name}" Knowledge/ ~/.cache/fit/outpost/
 ````
 
 Expected: no matches except this erasure report.

package/templates/.claude/skills/req-scan/SKILL.md

@@ -3,7 +3,7 @@ name: req-scan
 description: >
   Scan publicly available sources for candidates who indicate they are open for
   hire. Uses WebFetch to read public APIs (HN Algolia, GitHub, dev.to).
-  Writes prospect notes to knowledge/Prospects/. Maintains
+  Writes prospect notes to Knowledge/Prospects/. Maintains
   cursor/dedup state in ~/.cache/fit/outpost/head-hunter/. Use when the
   head-hunter agent is woken or when the user asks to scan for open candidates.
 ---
@@ -34,7 +34,7 @@ updates.
 
 ## Outputs
 
-- `knowledge/Prospects/{Name}.md`.
+- `Knowledge/Prospects/{Name}.md`.
 - Updated `cursor.tsv`, `seen.tsv`, `prospects.tsv`, and `log.md`.
 - `~/.cache/fit/outpost/state/head_hunter_triage.md`.
 
@@ -103,7 +103,7 @@ Write notes for **strong** and **moderate** matches only, using the template in
 [references/template.md](references/template.md).
 
 ```bash
-mkdir -p "knowledge/Prospects"
+mkdir -p "Knowledge/Prospects"
 ```
 
 ### 5. Update state

package/templates/.claude/skills/req-scan/references/template.md

@@ -1,6 +1,6 @@
 # Prospect Note Template
 
-Reference for `req-scan` Step 4. Save to `knowledge/Prospects/{Name}.md`.
+Reference for `req-scan` Step 4. Save to `Knowledge/Prospects/{Name}.md`.
 
 ```markdown
 # {Name}

package/templates/.claude/skills/req-screen/SKILL.md

@@ -22,7 +22,7 @@ This is **Stage 1** of a three-stage hiring pipeline:
 
 ## Trigger
 
-- A new CV is added to `knowledge/Candidates/{Name}/`.
+- A new CV is added to `Knowledge/Candidates/{Name}/`.
 - A CV appears in `~/Downloads/` and is associated with a candidate.
 - The user asks to screen, evaluate, or assess a CV.
 - The user asks "is this person worth interviewing?".
@@ -35,16 +35,16 @@ This is **Stage 1** of a three-stage hiring pipeline:
 
 ## Inputs
 
-- CV file path (e.g. `knowledge/Candidates/{Name}/CV.pdf`).
+- CV file path (e.g. `Knowledge/Candidates/{Name}/CV.pdf`).
 - Target role (optional).
-- Existing `knowledge/Candidates/{Name}/brief.md`, if any.
-- `knowledge/Roles/*.md` matching the candidate's `Req` (provides `Level`,
+- Existing `Knowledge/Candidates/{Name}/brief.md`, if any.
+- `Knowledge/Roles/*.md` matching the candidate's `Req` (provides `Level`,
   `Discipline`, `Hiring manager`, `Domain lead`).
 
 ## Outputs
 
-- `knowledge/Candidates/{Name}/screening.md` — structured assessment.
-- Updated `knowledge/Candidates/{Name}/brief.md` — skills + summary enriched.
+- `Knowledge/Candidates/{Name}/screening.md` — structured assessment.
+- Updated `Knowledge/Candidates/{Name}/brief.md` — skills + summary enriched.
 
 <do_confirm_checklist goal="Verify the screening is grounded and
 decision-rule-compliant">
@@ -76,8 +76,8 @@ Extract the fields listed in
 If `brief.md` carries a `Req`, look up the matching Role file:
 
 ```bash
-ls knowledge/Roles/ | grep "{req_number}"
-cat "knowledge/Roles/{matching file}"
+ls Knowledge/Roles/ | grep "{req_number}"
+cat "Knowledge/Roles/{matching file}"
 ```
 
 Use the Role's `Level` and `Discipline` as the target unless the user specified
@@ -127,7 +127,7 @@ Pick the recommendation using the decision rules and threshold rule in
 
 ### 7. Write the screening
 
-Save to `knowledge/Candidates/{Name}/screening.md` using the template in
+Save to `Knowledge/Candidates/{Name}/screening.md` using the template in
 [references/template.md](references/template.md). Include the **Suggested
 Interview Questions** when the recommendation is "Interview" or "Interview with
 focus areas":

package/templates/.claude/skills/req-screen/references/template.md

@@ -1,7 +1,7 @@
 # Screening Output Template
 
 Reference template for `req-screen` Step 6. Save to
-`knowledge/Candidates/{Name}/screening.md`. Filename **must** be `screening.md`
+`Knowledge/Candidates/{Name}/screening.md`. Filename **must** be `screening.md`
 — not `assessment.md`, `cv-screening.md`, or any variant. If a misnamed
 screening file exists in the folder (look for `# CV Screening` in the header),
 delete it after writing.

package/templates/.claude/skills/req-track/SKILL.md

@@ -1,13 +1,13 @@
 ---
 name: req-track
-description: Scan synced email threads for recruitment candidates, extract structured profiles, and create/update notes in knowledge/Candidates/. Use when the user asks to track candidates, process recruitment emails, or update the hiring pipeline.
+description: Scan synced email threads for recruitment candidates, extract structured profiles, and create/update notes in Knowledge/Candidates/. Use when the user asks to track candidates, process recruitment emails, or update the hiring pipeline.
 ---
 
 # Track Candidates
 
 Scan synced email threads from `~/.cache/fit/outpost/apple_mail/` for
 recruitment candidates. Extract structured candidate profiles and create or
-update notes in `knowledge/Candidates/`. Builds a local, searchable recruitment
+update notes in `Knowledge/Candidates/`. Builds a local, searchable recruitment
 pipeline from scattered email threads.
 
 ## Trigger
@@ -29,7 +29,7 @@ pipeline from scattered email threads.
 - `~/.cache/fit/outpost/apple_mail/attachments/` — CV/resume attachments.
 - `~/.cache/fit/outpost/apple_calendar/*.json` — calendar events (for
   cross-source inference).
-- `knowledge/Roles/*.md` — open role/requisition files (metadata inheritance).
+- `Knowledge/Roles/*.md` — open role/requisition files (metadata inheritance).
 - `~/.cache/fit/outpost/state/graph_processed` — processed-file index (shared
   with `extract-entities`).
 - `~/.cache/fit/outpost/state/identity.md` — user identity for self-exclusion
@@ -37,10 +37,10 @@ pipeline from scattered email threads.
 
 ## Outputs
 
-- `knowledge/Candidates/{Full Name}/brief.md` — candidate profile note.
-- `knowledge/Candidates/{Full Name}/CV.pdf` (or `CV.docx`) — local CV copy.
-- `knowledge/Candidates/{Full Name}/headshot.jpeg` — candidate photo.
-- `knowledge/Roles/*.md` — created/updated role files (Candidates tables
+- `Knowledge/Candidates/{Full Name}/brief.md` — candidate profile note.
+- `Knowledge/Candidates/{Full Name}/CV.pdf` (or `CV.docx`) — local CV copy.
+- `Knowledge/Candidates/{Full Name}/headshot.jpeg` — candidate photo.
+- `Knowledge/Roles/*.md` — created/updated role files (Candidates tables
   rebuilt).
 - `~/.cache/fit/outpost/state/graph_processed` — updated with processed threads.
 
@@ -84,31 +84,31 @@ here.
 ### 2. Build candidate, people, and org indexes
 
 ```bash
-ls -d knowledge/Candidates/*/
+ls -d Knowledge/Candidates/*/
 ```
 
 Read each existing brief's header (Name, Role, Source, Status) to build a mental
-index. Also scan `knowledge/People/`, `knowledge/Organizations/`, and
-`knowledge/Projects/` to resolve recruiter names, agency orgs, and project
+index. Also scan `Knowledge/People/`, `Knowledge/Organizations/`, and
+`Knowledge/Projects/` to resolve recruiter names, agency orgs, and project
 links.
 
-### 3. Sync `knowledge/Roles/`
+### 3. Sync `Knowledge/Roles/`
 
 This keeps role metadata current and enables inheritance.
 
 1. Read each Role file's Info block to map Req → Role file path, Hiring manager,
    Domain lead, recruiter, Channel.
 2. Find Reqs referenced by candidate briefs but missing a Role file:
-   `rg "^\*\*Req:\*\*" knowledge/Candidates/*/brief.md`. For each missing Req,
+   `rg "^\*\*Req:\*\*" Knowledge/Candidates/*/brief.md`. For each missing Req,
    create a stub using the **Role file stub** in
    [references/templates.md](references/templates.md), then enrich by searching
-   the graph: `rg "{req_number}" knowledge/`.
+   the graph: `rg "{req_number}" Knowledge/`.
 3. Rebuild each Role file's `## Candidates` table by scanning briefs:
-   `rg -l "Req:.*{req_number}" knowledge/Candidates/*/brief.md`. Use the **Role
+   `rg -l "Req:.*{req_number}" Knowledge/Candidates/*/brief.md`. Use the **Role
    Candidates table** format from `references/templates.md`. Sort by First seen,
    newest first.
 4. If a Role file has a hiring manager but no domain lead, walk the
-   `**Reports to:**` chain in `knowledge/People/` to a VP or senior leader.
+   `**Reports to:**` chain in `Knowledge/People/` to a VP or senior leader.
 
 ### 4. Identify recruitment threads
 
@@ -137,7 +137,7 @@ forwarded headers).
 
 ### 8. Write or update the candidate note
 
-For **new** candidates, create `knowledge/Candidates/{Full Name}/brief.md` from
+For **new** candidates, create `Knowledge/Candidates/{Full Name}/brief.md` from
 the **Candidate brief** template in
 [references/templates.md](references/templates.md). Place the **Extra Info
 fields** after `Last activity` in the order shown there. Add the **Optional
@@ -149,7 +149,7 @@ known Info fields, and add new Skills.
 
 ### 9. Capture cross-candidate insights
 
-Update `knowledge/Candidates/Insights.md` only when an observation is
+Update `Knowledge/Candidates/Insights.md` only when an observation is
 high-signal: candidate may suit a **different role**, is a **strong match** for
 a specific team or leader, a meaningful **comparison between candidates**, or a
 hiring trade-off needs to be **remembered across sessions**.

package/templates/.claude/skills/req-track/references/fields.md

@@ -43,13 +43,13 @@ Field map and resolution rules for Step 2 of `req-track`.
 
 Stop at the first match:
 
-1. **Req-first inheritance** — look up `knowledge/Roles/*.md` for the matching
+1. **Req-first inheritance** — look up `Knowledge/Roles/*.md` for the matching
    Req; inherit Hiring manager and Domain lead from the Role file.
 2. **Calendar inference** —
    `rg -l "{Candidate Name}" ~/.cache/fit/outpost/apple_calendar/`. The non-user
    organizer of an interview event is likely the hiring manager.
 3. **Email inference** — internal To/CC recipients (besides the user)
-   cross-checked against `knowledge/People/`.
+   cross-checked against `Knowledge/People/`.
 4. **Reporting chain** — read the hiring manager's `**Reports to:**` field; walk
    up to a VP / senior leader for the domain lead.
 5. **Staffing project timeline** — search staffing notes for the candidate or
@@ -71,7 +71,7 @@ hiring decisions, assessment criteria, or candidate visibility.
 
 ## Source and recruiter
 
-- Map sender email domain to an organization in `knowledge/Organizations/`.
+- Map sender email domain to an organization in `Knowledge/Organizations/`.
 - The person who sent or forwarded the profile is the recruiter — link with
   `[[People/Name]]`.
 - Create the organization or recruiter notes if missing.
@@ -80,9 +80,9 @@ hiring decisions, assessment criteria, or candidate visibility.
 
 ```bash
 ls ~/.cache/fit/outpost/apple_mail/attachments/{thread_id}/
-mkdir -p "knowledge/Candidates/{Full Name}"
+mkdir -p "Knowledge/Candidates/{Full Name}"
 cp "~/.cache/fit/outpost/apple_mail/attachments/{thread_id}/{file}" \
-   "knowledge/Candidates/{Full Name}/CV.pdf"
+   "Knowledge/Candidates/{Full Name}/CV.pdf"
 ```
 
 Use `CV.pdf` (PDF) or `CV.docx` (Word). The `## CV` link is `./CV.pdf`.
@@ -106,9 +106,9 @@ Match by name fragment (case-insensitive). Prefer the Downloads match. Always
 write `headshot.jpeg`:
 
 ```bash
-cp "{src}" "knowledge/Candidates/{Full Name}/headshot.jpeg"
+cp "{src}" "Knowledge/Candidates/{Full Name}/headshot.jpeg"
 # PNG → JPEG
-magick "{src}.png" "knowledge/Candidates/{Full Name}/headshot.jpeg"
+magick "{src}.png" "Knowledge/Candidates/{Full Name}/headshot.jpeg"
 # HEIC → JPEG
-magick "{src}.heic" "knowledge/Candidates/{Full Name}/headshot.jpeg"
+magick "{src}.heic" "Knowledge/Candidates/{Full Name}/headshot.jpeg"
 ```

package/templates/.claude/skills/req-track/references/signals.md

@@ -10,7 +10,7 @@ no signal — most email is not.
 
 ## Recruiter sender domains
 
-Sender domain maps to an organization in `knowledge/Organizations/` tagged as a
+Sender domain maps to an organization in `Knowledge/Organizations/` tagged as a
 recruitment agency. When no agencies are catalogued yet, treat these patterns as
 hints:
 
@@ -38,5 +38,5 @@ Structured candidate descriptions containing:
 
 ## Follow-up on existing candidates
 
-A thread mentions a candidate already in `knowledge/Candidates/` by name —
+A thread mentions a candidate already in `Knowledge/Candidates/` by name —
 process it to update pipeline status.

package/templates/.claude/skills/req-track/references/templates.md

@@ -50,7 +50,7 @@ Sort by First seen, newest first.
 ## Candidate brief
 
 Used by Step 5 for new candidates. File:
-`knowledge/Candidates/{Full Name}/brief.md`.
+`Knowledge/Candidates/{Full Name}/brief.md`.
 
 ```markdown
 # {Full Name}

package/templates/.claude/skills/req-workday/SKILL.md

@@ -2,7 +2,7 @@
 name: req-workday
 description: >
   Import candidates from a Workday requisition export (.xlsx) into
-  knowledge/Candidates/. Parses requisition metadata and candidate data,
+  Knowledge/Candidates/. Parses requisition metadata and candidate data,
   creates candidate briefs and CV.md files from resume text, and integrates
   with the existing req-track pipeline. Use when the user provides a
   Workday export file or asks to import candidates from an XLSX requisition
@@ -12,7 +12,7 @@ description: >
 # Workday Requisition Import
 
 Import candidates from a Workday requisition export (`.xlsx`) into
-`knowledge/Candidates/`. Extract requisition metadata and candidate profiles,
+`Knowledge/Candidates/`. Extract requisition metadata and candidate profiles,
 create standardized briefs and `CV.md` files from the embedded resume text, and
 integrate with the `req-track` pipeline format.
 
@@ -36,9 +36,9 @@ integrate with the `req-track` pipeline format.
 
 ## Outputs
 
-- `knowledge/Candidates/{Clean Name}/brief.md` — candidate profile.
-- `knowledge/Candidates/{Clean Name}/CV.md` — resume text as markdown.
-- `knowledge/Roles/{Req ID} — {Title}.md` — created or updated.
+- `Knowledge/Candidates/{Clean Name}/brief.md` — candidate profile.
+- `Knowledge/Candidates/{Clean Name}/CV.md` — resume text as markdown.
+- `Knowledge/Roles/{Req ID} — {Title}.md` — created or updated.
 - Updated existing briefs when a candidate already exists.
 
 <do_confirm_checklist goal="Verify the Workday import is consistent with
@@ -46,7 +46,7 @@ req-track">
 
 - [ ] XLSX parsed; candidate count matches the parser summary.
 - [ ] Requisition metadata extracted (ID, title; HM/recruiter when available).
-- [ ] Each candidate has a directory under `knowledge/Candidates/{Clean Name}/`
+- [ ] Each candidate has a directory under `Knowledge/Candidates/{Clean Name}/`
       (annotation stripped).
 - [ ] `CV.md` created for every candidate with resume text — faithfully
       reproduced (no rewriting).
@@ -93,13 +93,13 @@ header indices, name annotations) are in
 ### 3. Create or update the Role file
 
 ```bash
-ls knowledge/Roles/ | grep "{Req ID}"
+ls Knowledge/Roles/ | grep "{Req ID}"
 ```
 
 Use the **Role file stub** in
 [references/templates.md](references/templates.md). Resolve the domain lead by:
 
-1. `rg "{Req ID}" knowledge/` — look in project timelines, People notes, Topics
+1. `rg "{Req ID}" Knowledge/` — look in project timelines, People notes, Topics
    for context.
 2. Reading the hiring manager's People note for `**Reports to:**` and walking up
    to a VP or senior leader.
@@ -111,7 +111,7 @@ If the Role file already exists, follow the existing-file rules in
 ### 4. Build the candidate index
 
 ```bash
-ls -d knowledge/Candidates/*/ 2>/dev/null
+ls -d Knowledge/Candidates/*/ 2>/dev/null
 ```
 
 Match imported candidates against existing notes by name (fuzzy — middle names,
@@ -126,7 +126,7 @@ step value in the Pipeline entry.
 ### 6. Write `CV.md`
 
 For every candidate with resume text, create
-`knowledge/Candidates/{Clean Name}/CV.md` using the **CV.md template** in
+`Knowledge/Candidates/{Clean Name}/CV.md` using the **CV.md template** in
 [references/templates.md](references/templates.md).
 
 ### 7. Write or update the brief
@@ -136,7 +136,7 @@ Brief layout (new candidates) and edit rules (existing candidates):
 [references/templates.md](references/templates.md).
 
 ```bash
-mkdir -p "knowledge/Candidates/{Clean Name}"
+mkdir -p "Knowledge/Candidates/{Clean Name}"
 ```
 
 For existing candidates, apply targeted Edit operations only — never rewrite the
@@ -145,7 +145,7 @@ file.
 ### 8. Capture insights
 
 After the batch, review for strategic observations and add bullets to
-`knowledge/Candidates/Insights.md` under `## Placement Notes` with
+`Knowledge/Candidates/Insights.md` under `## Placement Notes` with
 `[[Candidates/Name/brief|Name]]` links. See `req-track` Step 5b for the
 inclusion criteria.
 

package/templates/.claude/skills/req-workday/references/templates.md

@@ -5,7 +5,7 @@ Reference templates for `req-workday` Steps 1b, 4, and 5.
 ## Role file stub
 
 Use when no Role file exists for the requisition. Filename:
-`knowledge/Roles/{Req ID} — {Short Title}.md`.
+`Knowledge/Roles/{Req ID} — {Short Title}.md`.
 
 ```markdown
 # {Requisition Title}
@@ -41,7 +41,7 @@ provides one and the field is `—`; update `Last activity` to today; append
 
 ## CV.md template
 
-Save to `knowledge/Candidates/{Clean Name}/CV.md` when resume text exists. Skip
+Save to `Knowledge/Candidates/{Clean Name}/CV.md` when resume text exists. Skip
 when it doesn't.
 
 ```markdown
@@ -60,7 +60,7 @@ when it doesn't.
 
 ## Candidate brief
 
-Save to `knowledge/Candidates/{Clean Name}/brief.md`. Format follows
+Save to `Knowledge/Candidates/{Clean Name}/brief.md`. Format follows
 `req-track`.
 
 ```markdown

package/templates/.claude/skills/send-chat/SKILL.md

@@ -35,7 +35,7 @@ Run when the user asks to:
 When the user mentions ANY person:
 
 1. **STOP** — Do not open the chat platform yet
-2. **SEARCH** — Look them up: `rg -l "{name}" knowledge/People/`
+2. **SEARCH** — Look them up: `rg -l "{name}" Knowledge/People/`
 3. **READ** — Read their note to understand context, role, recent interactions
 4. **UNDERSTAND** — Know who they are, what you've been working on together
 5. **THEN PROCEED** — Only now compose the message and use browser automation
@@ -53,10 +53,10 @@ a full name using the knowledge graph:
 
 ```bash
 # Find person by partial name
-rg -l -i "{name}" knowledge/People/
+rg -l -i "{name}" Knowledge/People/
 
 # If ambiguous, read candidates to disambiguate
-cat "knowledge/People/{Candidate}.md"
+cat "Knowledge/People/{Candidate}.md"
 ```
 
 **If ambiguous** (multiple matches), ask the user which person they mean — list
@@ -73,7 +73,7 @@ can review and edit the exact message before it's sent.
 ### Draft Workflow
 
 1. **Compose the message** based on context and user intent.
-2. **Write it to a draft file** at `drafts/chat-{recipient-slug}-{date}.md`
+2. **Write it to a draft file** at `Drafts/chat-{recipient-slug}-{date}.md`
    - `{recipient-slug}` = lowercase, hyphenated full name (e.g. `sarah-chen`)
    - `{date}` = ISO date (e.g. `2026-02-19`)
 3. **Show the user the draft** — display the file path and contents.

package/templates/.claude/skills/upstream-instructions/SKILL.md

@@ -0,0 +1,154 @@
+---
+name: upstream-instructions
+description: Track changes made to this installation's instruction files — the root CLAUDE.md, agent profiles, and skills — and record them in a single root CHANGELOG.md so improvements can be contributed back to the upstream monorepo. Use when CLAUDE.md, agents, or skills have been modified, added, or removed locally and those changes should be documented for upstream.
+---
+
+# Upstream Instructions
+
+Track changes to this installation's **instructions** and record them in one
+root `CHANGELOG.md` so improvements can be contributed back to the upstream
+monorepo. "Instructions" means all three surfaces, treated equally:
+
+- **`CLAUDE.md`** (root) — installation-wide instructions.
+- **`.claude/agents/*.md`** — agent profiles.
+- **`.claude/skills/*/`** — skills (`SKILL.md` and reference files).
+
+## Trigger
+
+- The user asks to prepare local instruction changes for upstream contribution.
+- The root `CLAUDE.md`, an agent profile, or a skill has been modified, added,
+  or removed.
+- The user wants to document what changed locally before syncing upstream.
+
+## Prerequisites
+
+- A working Outpost installation with `CLAUDE.md`, `.claude/agents/`, and
+  `.claude/skills/`.
+- Git available for change detection.
+
+## Inputs
+
+- `CLAUDE.md` — root installation instructions.
+- `.claude/agents/*.md` — agent profiles.
+- `.claude/skills/*/SKILL.md` and reference files — skills.
+- `CHANGELOG.md` (root) — the existing changelog, for the last documented date.
+- Git history and working tree — change detection.
+
+## Outputs
+
+- `CHANGELOG.md` (root) — a **single** reverse-chronological changelog covering
+  all three surfaces. There are no per-skill or per-agent changelogs.
+
+<do_confirm_checklist goal="Verify the changelog is upstream-ready">
+
+- [ ] Exactly one root `CHANGELOG.md`; no stray per-skill / per-agent changelogs
+      left behind.
+- [ ] Every entry names its **Scope** (which surface(s) and file(s) it touched).
+- [ ] Each entry has **What**, **Why**, and **Details**.
+- [ ] Descriptions are specific enough for an upstream maintainer to act on (not
+      "updated CLAUDE.md" / "fixed stuff").
+- [ ] New skills/agents include a brief description of their purpose; removed
+      ones explain why.
+- [ ] Dates come from git history (or the date the change was made for
+      uncommitted edits), not guessed.
+- [ ] No duplicate entries for the same change.
+
+</do_confirm_checklist>
+
+## Procedure
+
+### 1. Find the last documented state
+
+```bash
+head -20 CHANGELOG.md 2>/dev/null   # newest date already recorded, if any
+```
+
+### 2. Identify changed instructions
+
+Detect changes across all three surfaces — committed and uncommitted:
+
+```bash
+# Committed changes since the last documented date (or all history)
+git log --after="<last-entry-date>" --name-status -- \
+  CLAUDE.md '.claude/agents/' '.claude/skills/'
+
+# Uncommitted working-tree changes (common in a live installation)
+git status --short -- CLAUDE.md '.claude/agents/' '.claude/skills/'
+```
+
+If no `CHANGELOG.md` exists yet, consider all changes since the initial commit.
+
+### 3. Classify each change
+
+| Surface       | What it covers                                  |
+| ------------- | ----------------------------------------------- |
+| `CLAUDE.md`   | Root installation instructions                  |
+| `agent:<name>`| A profile in `.claude/agents/`                  |
+| `skill:<name>`| A skill in `.claude/skills/`                    |
+
+| Type       | Description                                            |
+| ---------- | ------------------------------------------------------ |
+| `added`    | New instruction file that doesn't exist upstream       |
+| `modified` | Existing instruction updated                           |
+| `removed`  | Instruction file or directory deleted                  |
+| `renamed`  | File or directory renamed                              |
+
+A single change often spans surfaces (e.g. a KB-structure change touching
+`CLAUDE.md`, several agents, and several skills) — record it as **one entry**
+whose Scope lists every surface touched. Diff to confirm the actual change:
+
+```bash
+git diff <commit> -- CLAUDE.md
+git diff <commit> -- '.claude/agents/<agent>.md'
+git diff <commit> -- '.claude/skills/<skill>/'
+```
+
+### 4. Describe each change
+
+Every entry must answer:
+
+1. **What changed?** — the specific instruction or behaviour modified.
+2. **Why?** — the problem encountered or improvement discovered in use.
+3. **Details** — a summary of the actual change (not a full diff).
+
+Good: "Agents now read `Knowledge/Priorities/` at the start of every wake and
+flag anything that threatens a priority — drafts and triage were ignoring
+strategic context."
+
+Bad: "Updated the agents" / "Fixed stuff" / "Changed line 42".
+
+### 5. Write the changelog
+
+Create or update the single root `CHANGELOG.md` (newest first):
+
+```markdown
+# Outpost Instructions Changelog
+
+Changes to this installation's instructions — root `CLAUDE.md`, agent profiles
+(`.claude/agents/`), and skills (`.claude/skills/`) — for consideration upstream
+in the Forward Impact monorepo. Documents only; pushes nothing.
+
+## <YYYY-MM-DD>
+
+**Scope:** <surfaces/files, e.g. "skill: extract-entities; CLAUDE.md; agents: librarian, recruiter">
+**Type:** <added | modified | removed | renamed>
+
+**What:** <one-line summary>
+
+**Why:** <problem or improvement that motivated it>
+
+**Details:**
+<2–6 lines describing the specific changes across the listed surfaces>
+
+---
+```
+
+Worked examples in [references/examples.md](references/examples.md).
+
+## Notes
+
+- This skill **documents only** — it does not push or merge anything.
+- The single root `CHANGELOG.md` is consumed by the **downstream-instructions**
+  skill in the upstream monorepo.
+- When in doubt about whether a change is upstream-worthy, include it; the
+  upstream maintainer decides what to incorporate.