bindify
Use when the user references wip-docs/, .bindify/, bindify commands, plan/proposal/update/verify artifacts, coordinator workflows, multi-agent execution with shared markdown state, iterate-planning-mode, prototype-mode, verify-worktree, drafting a brief, investigating a feature or fix, looking up prior fixes, or when bindify chat/planning/implementation should follow caveman and ponytail.
How do I install this agent skill?
npx skills add https://github.com/den-ree/blanche-skills --skill bindifyIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The bindify skill is a comprehensive project management and automation framework for AI agents. It facilitates software development workflows by managing documentation (WIP docs), architecture maps, and git operations. It uses a dedicated GitHub App ('Blanche') for pull request automation, requiring a private key (PEM) provided via environment variables. The skill ships several local Python and Bash scripts to handle token minting, plan parsing, and worktree management. Analysis confirms that these scripts follow security best practices, such as using restricted file permissions for credentials and avoiding shell injection in command execution. No malicious patterns or security risks were identified.
- Socketpass
No alerts
- Snykwarn
Risk: MEDIUM · 1 issue
What does this agent skill do?
Bindify
Filesystem protocol for AI agents on software work. Active plan docs live in wip-docs/<feature>/<plan>/ at the project root. The tracking repo (.bindify/ or bindify/) holds architecture, migrated plans, and history/.
Load command and template files from this skill's references/ on demand. Layout detail: references/docs/layout.md. Pipeline diagrams: references/docs/workflow.md.
Three path layers
| Layer | Where | Use |
|---|---|---|
| Skill | This install | Commands + templates |
| Active WIP | wip-docs/<feature>/<plan>/ | Write brief/proposal/plan/updates/verify. Coordinator at wip-docs/<feature>/coordinator.md |
| Tracking | .bindify/ or bindify/ | Read project/architecture; migrate development/ at packaging; history may start at draft PR |
Never open tracking looking for skill commands or templates.
Read by role
| Role | Read |
|---|---|
| Planner | working-style.md, draft-brief, generate-proposal, save-agent-plan, publish-plan (branch stack), wait for ready |
| Executor | working-style.md, iterate-planning-mode, summarize-work-for-updates, next_step.py, generate-verify on finish |
| Reviewer | review-plan (build/CI gate + one fixes round) |
| Packager | log-pr, publish-bindify-pr, generate-history-summary |
When to invoke each command
| User signal | Command | Reference |
|---|---|---|
| adapt / configure project | configure-project | references/commands/configure-project.md |
| coordinate updates | coordinate-updates | references/commands/coordinate-updates.md |
| capture brief | draft-brief | references/commands/draft-brief.md |
| generate proposal | generate-proposal | references/commands/generate-proposal.md |
| save plan / proposal approved | save-agent-plan | references/commands/save-agent-plan.md |
| /bindify / publish plan | publish-plan | references/commands/publish-plan.md |
| run step / iterate | iterate-planning-mode | references/commands/iterate-planning-mode.md |
| review open plan PR | review-plan | references/commands/review-plan.md |
| prototype | prototype-mode | references/commands/prototype-mode.md |
| verify worktree (macOS) | verify-worktree | references/commands/verify-worktree.md |
| after a step | summarize-work-for-updates | references/commands/summarize-work-for-updates.md |
| scan architecture | scan-architecture | references/commands/scan-architecture.md |
| log PR | log-pr | references/commands/log-pr.md |
| history summary / abandoned | generate-history-summary | references/commands/generate-history-summary.md |
| regenerate overview | regenerate-overview | references/commands/regenerate-overview.md |
| publish bindify PR / package | publish-bindify-pr | references/commands/publish-bindify-pr.md |
| generate verify | generate-verify | references/commands/generate-verify.md |
| investigate / prior fixes | investigate | references/commands/investigate.md |
| research codebase | research-codebase | references/commands/research-codebase.md |
| repair links | update-links | references/commands/update-links.md |
| log hotfix | log-hotfix | references/commands/log-hotfix.md |
| abandon / drop plan | abandon-plan | references/commands/abandon-plan.md |
Hard rules
- Active docs stay in
wip-docs/<feature>/<plan>/until packaging. Coordinator iswip-docs/<feature>/coordinator.md. Several plans of one feature can sit side by side. Read sibling plan docs before writing a proposal or plan, and again before a step. Do not write active plan artifacts into trackingdevelopment/during apply. Flat files directly inwip-docs/mean stop and ask. - Branch stack: see
publish-plan.md. A different plan, while on another plan orplan-fixesbranch of the same feature, is cut from the current tip. Its draft PR still targetsfeature/<feature-slug>. Merge that PR with a merge commit. Executors never create branches. Onlyreview-plancreatesplan-fixes/. - One history card per plan. Slug is
<feature-slug>-<plan-slug>. Iterations andplan-fixesrefresh that card. - Packaging migrates and deletes that plan folder only. The feature folder goes away when no plan folders remain.
- Never modify existing plan steps during apply. Execution state →
updates.md.review-planmay append steps onplan-fixes/only. - Every update entry needs Impact & Connections + Architecture. Append-only for
updates.mdandhotfixes.md. - Architecture
Responsibilityis human-owned.scan-architecture fillappends change-log lines and edges only — and runs at packaging, not per step. - Package per plan on the plan head after sign-off, before the plan PR merges into its base. Feature → main is human, outside bindify. History
merged= plan PR into its base. - Brief ≤ 150 words. Proposal: Primer ≤ 300 words (+ table); Risks ≤ 3 bullets. Primer teaches the technology; project choices are facts in
In this project. - Prototype-first for user-facing features; Prototype steps carry a human checkpoint.
- Repo-relative paths only. No code dumps in context files.
- Investigate tracking names/fixes before product code. Clarify before
brief.md. - Chat is caveman; artifacts stay normal prose. Plan/build with ponytail (
references/docs/working-style.md). - Blanche owns agent PRs: commits
--author="blanche <blanche@bindify.app>";gh pr createviablanche_gh_token.pyso author isblanche[bot]. Missing credentials → stop. Seereferences/docs/blanche-github-app.md. update-linksruns atgenerate-verifyand at migration (plus history/hotfix/architecture writes) — not after every WIP write.
Two responsibilities
| Phase | Document | Who |
|---|---|---|
| plan / propose / review | brief → proposal | human + AI |
| apply | plan + updates | executor via iterate |
| verify | verify.md | human checklist |
| review | ready plan PR | reviewer via review-plan |
| package | migrate + history | packager |
| abandon | history abandoned | human-triggered abandon-plan |
What to read next
references/AGENTS.md— drop-in entry for host reposreferences/docs/working-style.md— clarify, caveman, ponytailreferences/docs/workflow.md— branch stack visuals, rolesreferences/docs/layout.md— folder treesreferences/docs/blanche-github-app.md— App token setupreferences/commands/<name>.md— load before invokingreferences/templates/<name>.md— when creating artifacts
When invoking a command, read that command file first.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/den-ree/blanche-skills/bindify">View bindify on skillZs</a>