feat: adopt ACF in artemisa and homedir for testing

[VECTORG99]

Summary

Adopt ACF in artemisa (to automate the full SDLC pipeline and put it to the test) and in homedir (to put it to the test in the existing autonomous SDLC pipeline). This issue tracks the adoption plan, validation criteria, and lessons learned — all work happens in those repos' own issues/PRs, this issue is the coordination point in ACF.

Context

• ACF repo: https://github.com/VECTORG99/acf
• Architecture: docs/ARCHITECTURE.md
• Compaction research: docs/COMPACTION.md
• Build process: docs/PROCESS.md
• ACF is inspired by homedir's SDLC pipeline but does NOT touch it — this issue proposes the reverse: homedir adopts ACF as a skill
• artemisa: automate the full ACF pipeline (context-load → stack-audit → issue-craft → pr-context → frontend-preview → launch) with compaction/caveman modes active

Affected Repos (work happens there, not here)

• artemisa — install ACF, automate issue/PR creation, test compaction + caveman
• homedir — install ACF alongside existing scc-* pipeline, test stack-audit + label-metadata

Constraint: This issue lives in the ACF repo only. No changes are made to artemisa or homedir from this repo. Each adoption is tracked in a separate issue opened in the respective repo.

Acceptance Criteria

• ACF installed in artemisa (.devin/skills/ or equivalent) — cp -r .devin/skills/* artemisa/.devin/skills/
• ACF installed in homedir (.devin/skills/ or equivalent) — cp -r .devin/skills/* homedir/.devin/skills/
• artemisa: at least 1 issue crafted via ACF with all required labels (type + priority + area)
• artemisa: at least 1 PR crafted via ACF with Closes #N and AC checkboxes
• artemisa: compaction (phase 7) triggered at least once, compacted snapshot verified
• artemisa: caveman (phase 8) tested at least once, caveman snapshot verified (<500 tokens)
• homedir: stack-audit (phase 2) run, orphan PRs / close gaps / escaped issues detected
• homedir: label-metadata (phase 6) labels created in the repo
• Both repos: document lessons learned and gaps found in ACF
• Feed lessons learned back into ACF via follow-up issues in this repo

Validation

• gh issue list --repo VECTORG99/acf --label "enhancement" shows this issue
• gh label list --repo artemisa-org/artemisa includes ACF canonical labels
• gh label list --repo VECTORG99/homedir includes ACF canonical labels
• artemisa and homedir git status clean before and after ACF installation (ACF is additive)

Complexity

Medium — 2 repos, 2 separate installations, but each is a cp -r + label setup + test run.

Notes

• ACF does NOT modify artemisa or homedir code — it installs as a skill directory
• The adoption is reversible: rm -rf .devin/skills/acf .devin/skills/0* removes ACF completely
• Compaction and caveman modes are the key innovations to test — they should reduce token usage in both repos' autonomous pipelines
• homedir already has an SDLC pipeline; ACF should complement it, not replace it. The stack-audit and label-metadata phases are the most valuable for homedir
• artemisa is the greenfield test — the full pipeline (including frontend-preview if applicable) should be exercised

Related

• ACF repo: https://github.com/VECTORG99/acf
• ACF architecture: docs/ARCHITECTURE.md
• ACF compaction research: docs/COMPACTION.md
• ACF build process: docs/PROCESS.md