Claim a Forge issue and then PROVE you hold the live lease before you touch it, using `forge issue owns <id>` (exit 0 iff you hold the single unexpired lease). Use this whenever winning the claim matters: right after `forge claim`, before you `dev`/edit/`close`/`release` a claimed issue, when two agents or a subagent fan-out contend for the same work, or before any irreversible step. A claim's `ok:true` does NOT prove you won — duplicate replays return it and expired leases get reclaimed; only `owns` proves it, so re-verify before close/release. Trigger on "claim this issue safely", "did I actually win the lease", "verify/prove ownership", "claim conflict", "two agents grabbed the same issue", "check I still own it before closing", or before ANY mutation of a claimed issue. NOT for plain single-issue create/update/close/comment with no ownership question (that is issue-basics), and NOT for read-only selecting or ranking the next ready issue without claiming (that is triage-ready).
Forge DEV stage — implement an already-planned /plan task list into committed, test-backed code. Reads tasks.md + plan.md, then drives each task through a subagent TDD loop (implementer → spec-compliance reviewer → code-quality reviewer) with RED-GREEN-REFACTOR, HARD-GATE evidence checks, and a spec-gap decision score. Use when a plan and task list already exist and it is time to implement — triggers: "/dev", "start the dev stage", "build the tasks", "implement tasks.md test-first", "run the implementer/reviewer TDD loop", "write the code for the planned tasks one by one", "work through the task list with subagents". Per-task coding ONLY. Do NOT use for creating the design doc or task list (that is plan), for the post-build type-check/lint/security/test gate (validate), for pushing the branch or opening the PR (ship), for addressing PR review feedback (review), or for orchestrating several stages / taking an issue end-to-end to a merged PR (smith).
Hermes⇄Forge boundary: Hermes CONSUMES Forge state, never a second source of truth. Use whenever a Hermes session runs on a Forge repo: at session start, before acting on an issue, or when you need CURRENT state. Read state ONLY via `forge orient` / `forge recap <issue-id>` (bounded JSON envelope), citing each source's `path`/`authority`; never reconstruct it from raw stores or kernel internals. Writeback ONLY via `forge comment`/`update`/`create`; NEVER leak Hermes profile or session memory into Forge state. Triggers: "orient me / current project state", "where did this fact come from, cite it", "orient came back truncated", "persist a decision into Forge", "safe to store in kernel state?". NOT the Forge session router over stage skills (kernel), NOT the human "what stage am I in" report (status), NOT everyday issue create/update/close/search CRUD (issue-basics), NOT ranking the next ready issue (triage-ready), NOT live PR monitoring (shepherd—outside Hermes, unseen by orient).
Everyday single-issue CRUD over the `forge issue` verbs: create/update/show/list/search/close/reopen/comment, set priority/labels/assignee, claim or release one issue, add/remove dependency edges, plus backlog `stats`. Use for ANY routine one-off issue op: "create an issue/bug/task for X", "update/edit issue <id>", "close or reopen this issue", "comment a handoff note on <id>", "list/filter open bugs by status/label/priority", "bump this to P1", "reassign to alice", "mark <id> blocked by <id>". Also the parity floor migrating off a Beads-style tracker (label/reopen/delete map to forge equivalents). Single-operation plumbing only. Does NOT choose, rank, or explain the next issue to work on or why it's blocked (use triage-ready); does NOT run claim-then-prove-lease-ownership safety (use claim-safety); does NOT drive an issue through the plan->dev->validate->ship pipeline or open a PR (use smith or stage skills); does NOT report the current stage or what's in flight (use status).
Forge kernel — umbrella index/router for a Forge project. Reach for this FIRST when orienting rather than executing: at session start, when you need the map of how the system fits together, or when unsure WHICH skill or `forge` verb a task belongs to. It indexes `smith` (end-to-end orchestrator), the stage ladder (plan → dev → validate → ship → review → verify), the utility/issue skills, and the day-to-day `forge` CLI verbs. Trigger on "how does the Forge workflow work", "which forge command or skill for X", "I'm new here, how is Forge set up", or "should this be a Forge issue or a TodoWrite". Index layer only — hand off the doing: rank/pick the next ready issue → `triage-ready`; current stage / "where am I" / active or stale work → `status`; create/update/close/search one issue → `issue-basics`; claim-then-prove ownership before mutating → `claim-safety`; drive one issue from plan to a merged PR under gates → `smith`; token-bounded state for the Hermes harness → `hermes-forge`.
Capture and retrieve durable project memory the right way -- and, when the graph backend is enabled, use temporal knowledge-graph memory. Reach for this whenever you are about to write down a lasting fact, decision, convention, or gotcha ("remember that...", "note for later", "save this"), or when you need to recall what was learned before ("what did we decide about X", "have we seen this"). It explains WHEN to use `forge remember` / `forge recall` versus a per-issue `forge issue comment`, how the two public backends work (local JSONL default, opt-in Graphiti; kernel is internal-only), and -- when Graphiti is enabled -- how to add episodes and search facts via the graphiti-memory MCP tools (add_memory, search_memory_facts, search_nodes) with group_id scoping and provenance. The local backend is always the offline floor. Not for: issue create/list/close ops (issue-basics), workflow status (status), or transient scratch notes.
Heavyweight EXTERNAL research reports — market, industry, competitive, strategic — via Parallel AI's paid pro/ultra processors that crawl and synthesize dozens of sources into one long cited report. Use when the user wants a market analysis, competitive landscape, industry deep-dive, multi-vendor/technology comparison, strategic recommendations, or market sizing/growth outlook — any multi-source synthesis needing more than 3-4 web searches. Typical phrasings: "market analysis of X", "competitive landscape comparing A/B/C", "industry deep-dive on...", "deep research report on...", "strategic report with predictions for 2027". WRONG tool for quick facts, a single-page fetch, one-URL scraping, or small structured-field extraction — use built-in WebSearch/WebFetch. Also NOT the Forge RESEARCH or PLAN stage: "run the research stage", "do Phase 2", or codebase/OWASP/DRY investigation into a design doc route to `research` and `plan` (they may INVOKE this skill). Requires PARALLEL_API_KEY.
Forge PLAN stage — first stop when starting a NEW or unscoped feature. Sets up an isolated worktree up front, then runs one-question-at-a-time brainstorming for design intent, commits a design doc, does technical/OWASP/DRY + codebase research, and produces a TDD task list for /dev. Trigger on "let's plan X", "scope a new feature", "brainstorm before we build", "write a design doc", "break this into tasks", or "set up a worktree and task list before coding". Reach for this even when the ask sounds like only design or only scoping — plan owns intent → research → task-list setup as one stage. NOT for driving a feature to a merged PR (that is smith), NOT for implementing tasks that already exist (dev), NOT for a standalone deep-research pass into an approved design doc (research), NOT for reporting where work stands or what is stale (status), and NOT for everyday issue create/list/close or picking the next ready issue (issue-basics / triage-ready).