feat(library): add marketing job with competitive content analysis

library/jobs/README.md

@@ -17,6 +17,7 @@ This walks you through configuring `DEEPWORK_ADDITIONAL_JOBS_FOLDERS` so the Dee
 | Job | Description |
 |-----|-------------|
 | [Engineer](./engineer) | Domain-agnostic engineering execution from product issue through PR merge and product sync, with TDD discipline |
+| [Marketing](./marketing) | Competitive analysis, content strategy, and social media planning with shared steps across workflows |
 | [Research](./research) | Multi-workflow research suite — deep investigation, quick summaries, material ingestion, and reproduction planning |
 | [Platform Engineer](./platform_engineer) | Incident response, observability, CI/CD, releases, security, cost management, and infrastructure |
 | [Repo](./repo) | Audit and configure repositories — labels, branch protection, milestones, and boards |
@@ -61,6 +62,25 @@ The job library provides:
 - **Templates**: Copy and adapt jobs for your own use cases
 - **Learning**: Understand the job definition format through real examples
 
+## Output conventions for library jobs
+
+Library jobs that produce durable outputs (reports, research, decisions) should defer to the user's environment for note creation mechanics rather than embedding specific tools or paths.
+
+**Pattern**: Signal the intent, not the mechanism.
+
+- **Do**: "The final report and supporting research are durable notes. If the user has a notes directory, follow its AGENTS.md conventions."
+- **Don't**: Embed `zk new reports/ --title ...` commands, specific frontmatter schemas, or directory paths in the job's `common_job_info`.
+
+**Why**: Users have different note-taking setups (zk, Obsidian, plain markdown, no notes repo). The user's notes directory AGENTS.md and their platform conventions (if any) are the source of truth for how to create notes. Library jobs that embed one system's mechanics become non-portable.
+
+**How it works in practice**:
+
+1. The job's `discover_context` step checks `NOTES_DIR` or `ZK_NOTEBOOK_DIR` and reads the AGENTS.md there.
+2. The agent's platform conventions (e.g., `process.notes`, `tool.zk-notes`) provide the mechanics for note creation.
+3. If no notes directory exists, the job falls back to `.deepwork/tmp/` with plain markdown.
+
+This separation means the same library job works for a Keystone user with a zk notebook, an Obsidian user, or someone with no notes system at all.
+
 ## Structure
 
 Each job in this library follows the same structure as the `.deepwork/jobs` subfolders in your local project:
@@ -76,6 +96,10 @@ library/jobs/
 │   ├── CLAUDE.md -> AGENTS.md
 │   ├── requirements.md      # RFC 2119 requirements specification
 │   └── steps/               # Step instruction files (also inlined in job.yml)
+├── marketing/               # Competitive content analysis workflows
+│   ├── job.yml
+│   ├── AGENTS.md            # Agent context and learnings
+│   └── CLAUDE.md -> AGENTS.md
 ├── platform_engineer/       # Platform engineering workflows
 │   ├── job.yml
 │   ├── AGENTS.md            # Agent context and learnings

library/jobs/marketing/AGENTS.md

@@ -0,0 +1,76 @@
+# Project context for marketing
+
+This is a **library job** -- reusable marketing workflows that any project can adopt.
+
+## Purpose
+
+General-purpose marketing workflows for competitive analysis, content strategy,
+social media planning, and campaign execution. Steps prefixed `cca_` are specific
+to competitive content analysis; prefixed `sm_` are specific to social media.
+Unprefixed steps (`discover_context`, `audit_own_content`, `create_campaign`,
+`brainstorm_content`) are shared across workflows.
+
+## Location
+
+This job lives in `library/jobs/marketing/`. It is NOT auto-installed -- users adopt it
+via `DEEPWORK_ADDITIONAL_JOBS_FOLDERS`, the `shared_jobs` workflow, or by copying it
+into their project's `.deepwork/jobs/` directory.
+
+## Workflows
+
+| Workflow                       | Steps                                                                                                      | Purpose                                                  |
+|--------------------------------|------------------------------------------------------------------------------------------------------------|----------------------------------------------------------|
+| `competitive_content_analysis` | discover_context, cca_research_all_competitors, cca_analyze_patterns, audit_own_content, cca_gap_analysis, cca_final_report, create_campaign, brainstorm_content | Competitive content analysis with own-content audit, campaign tracking, and brainstorm |
+| `social_media_setup`           | discover_context, audit_own_content, sm_define_strategy, create_campaign                                   | Content strategy and social media planning for a project |
+
+Per-competitor research is delegated to the shared `research` job (`research/quick` or `research/research`) via nested workflow calls. No marketing-internal research sub-workflow needed.
+
+## Planned workflows (not yet implemented)
+
+- `campaign_planning` -- strategy through execution
+- `content_calendar` -- content calendar across channels
+- `seo_audit` -- technical and content SEO audit
+- `email_marketing` -- campaigns and automations
+- `launch_comms` -- launch communications coordination
+- `analytics_attribution` -- performance and attribution models
+
+## Shared steps
+
+These steps are reused across sibling workflows:
+
+- **discover_context**: Project/brand context discovery with output directory inference
+- **audit_own_content**: Crawl and catalog existing content across all channels
+- **create_campaign**: Create milestone + parent issue on the user's project tracker
+- **brainstorm_content**: Interactive content ideation posted as issue comment
+
+## Output conventions
+
+Workflow outputs are durable notes. If the user has a notes directory
+(`NOTES_DIR` or `ZK_NOTEBOOK_DIR`), follow its AGENTS.md conventions for note
+creation and placement. The final report is a report-type note; supporting
+per-competitor research goes as literature notes linked via wikilinks. Fall back
+to `.deepwork/tmp/` only if no notes directory exists.
+
+This job does not embed note creation mechanics (zk commands, frontmatter
+format, tag conventions) because those vary by user setup. The user's notes
+repo AGENTS.md and the agent's platform conventions are the source of truth for
+how to create notes. See `process.notes` and `tool.zk-notes` for the canonical
+rules if available in the agent's instruction context.
+
+## Quality review learnings
+
+### 2026-04-01: Plant Caravan competitive content analysis
+
+- The workflow originally ended at `final_report`. Analysis without action tracking
+  left the user to manually create issues and milestones. Added `create_campaign`
+  and `brainstorm_content` steps to close the loop.
+- The user had specific content ideas they wanted to explore but the workflow gave
+  no opportunity for interactive brainstorming. The new `brainstorm_content` step
+  asks structured questions to capture the user's vision alongside the data-driven
+  recommendations.
+- Forgejo required `-r` flag on `yq` for token extraction — tea config stores
+  tokens as YAML strings that include quotes without raw output mode.
+- The gap analysis compared competitor content against a one-line summary of
+  the brand's content from context_brief.md rather than an actual audit. Added
+  `audit_own_content` step that crawls the brand's website, blog, and social
+  channels to produce a real inventory before gap analysis runs.

library/jobs/marketing/CLAUDE.md

@@ -0,0 +1 @@
+AGENTS.md