Using TermCanvas
Route first. Choose the lightest path that preserves correctness.
Routing
- If the user wants to challenge, stress-test, or critically review an idea, argument, or proposal, use
challenge.
- If the user asks to investigate a bug, debug, or diagnose an issue, use
investigate.
- If the user asks for a security review or audit, use
security-audit.
- If the user asks for a code review or diff review, use
code-review.
- If the user asks to test a site, QA a page, or verify a deploy, use
qa.
- If the task is simple, local, high-certainty, or faster in the current
agent, do it directly. Do not invoke Hydra by default.
- If the task needs an isolated worktree, file evidence, retry/status control,
or a staged workflow, use
hydra.
- Before using Hydra in a repo, ensure the project has current Hydra
instructions via
hydra init-repo or the TermCanvas Hydra enable action.
Hydra workflow patterns
Lead-driven, decision-point oriented. The Lead reads the codebase, picks
the strategy, and dispatches workers for the steps that need a fresh
agent process. Roles available: lead (the decider — not itself dispatched),
dev (writes code AND its tests), reviewer (independent cross-model
check). No separate researcher — the Lead does its own research. No
separate tester — dev owns its own test surface.
hydra init --intent "..." --repo . then dispatch dev -> reviewer
for ambiguous, risky, or PRD-driven work
- call
hydra watch after each dispatch to wait for the decision point
hydra spawn --task "..." --repo . for a single isolated worker
- use when the task split is already known and you do not need the
full Lead-driven loop
Hydra worker primitive
hydra spawn --task "..." --repo .
- one direct isolated worker terminal
- use when the task split is already known and you do not need a full workflow
Guardrails
- Do not describe Hydra workflows as automatic parallelism unless multiple
spawned workers are actually involved.
- When launching Claude/Codex tasks via TermCanvas CLI, use
termcanvas terminal create --prompt "..." rather than termcanvas terminal input.
- After
hydra dispatch, immediately start hydra watch — do not ask whether to watch.
- Use
hydra watch / hydra status / hydra ledger / hydra list --workflows
for workflows created by hydra init.
- Use
hydra list and hydra cleanup <agentId> for direct workers created by
hydra spawn.
- Prefer structured Hydra state and files over terminal prose.
Memory Graph
When the session context contains a <memory-graph> block from TermCanvas:
- Check "References" before reading a memory file — referenced files are likely also relevant, follow the links
- If a memory is marked "Time-sensitive" with a date that has clearly passed, verify its content against current project state before acting on it
- Do not cite memory-graph metadata to the user — it's for your navigation, not for display
1---2name: using-termcanvas3description: Use when starting work in a TermCanvas-managed repo to route between direct work, Hydra, or a narrow TermCanvas skill.4---56# Using TermCanvas78Route first. Choose the lightest path that preserves correctness.910## Routing1112- If the user wants to challenge, stress-test, or critically review an idea, argument, or proposal, use `challenge`.13- If the user asks to investigate a bug, debug, or diagnose an issue, use `investigate`.14- If the user asks for a security review or audit, use `security-audit`.15- If the user asks for a code review or diff review, use `code-review`.16- If the user asks to test a site, QA a page, or verify a deploy, use `qa`.17- If the task is simple, local, high-certainty, or faster in the current18 agent, do it directly. Do not invoke Hydra by default.19- If the task needs an isolated worktree, file evidence, retry/status control,20 or a staged workflow, use `hydra`.21- Before using Hydra in a repo, ensure the project has current Hydra22 instructions via `hydra init-repo` or the TermCanvas Hydra enable action.2324## Hydra workflow patterns2526Lead-driven, decision-point oriented. The Lead reads the codebase, picks27the strategy, and dispatches workers for the steps that need a fresh28agent process. Roles available: `lead` (the decider — not itself dispatched),29`dev` (writes code AND its tests), `reviewer` (independent cross-model30check). No separate researcher — the Lead does its own research. No31separate tester — dev owns its own test surface.3233- `hydra init --intent "..." --repo .` then dispatch `dev` -> `reviewer`34 for ambiguous, risky, or PRD-driven work35 - call `hydra watch` after each dispatch to wait for the decision point36- `hydra spawn --task "..." --repo .` for a single isolated worker37 - use when the task split is already known and you do not need the38 full Lead-driven loop3940## Hydra worker primitive4142- `hydra spawn --task "..." --repo .`43 - one direct isolated worker terminal44 - use when the task split is already known and you do not need a full workflow4546## Guardrails4748- Do not describe Hydra workflows as automatic parallelism unless multiple49 spawned workers are actually involved.50- When launching Claude/Codex tasks via TermCanvas CLI, use51 `termcanvas terminal create --prompt "..."` rather than `termcanvas terminal input`.52- After `hydra dispatch`, immediately start `hydra watch` — do not ask whether to watch.53- Use `hydra watch` / `hydra status` / `hydra ledger` / `hydra list --workflows`54 for workflows created by `hydra init`.55- Use `hydra list` and `hydra cleanup <agentId>` for direct workers created by56 `hydra spawn`.57- Prefer structured Hydra state and files over terminal prose.5859## Memory Graph6061When the session context contains a `<memory-graph>` block from TermCanvas:6263- Check "References" before reading a memory file — referenced files are likely also relevant, follow the links64- If a memory is marked "Time-sensitive" with a date that has clearly passed, verify its content against current project state before acting on it65- Do not cite memory-graph metadata to the user — it's for your navigation, not for display