Reported issues for heliograph
Pod holds 5 of 5 GitHub reports that passed its relevance review. This can include external user reports, maintainer-confirmed bugs, and concrete feature gaps. Treat them as evidence to inspect, not a count of distinct defects.
Back to heliograph.
Most discussed
fix: manifests send users to heliograph.io, and a test pins the old host
Four shipped manifests still name heliograph.dbhq.uk as the homepage or documentation. That host's apex now 301s to heliograph.io, the commercial site, so a user who clicks through from the plugin, the MCP registry, npm or the .mcpb bundle lands on a product page instead of the toolkit docs. The SKILL.md links and one CLI message reach docs.heliograph.io, but only through a redirect. And the coherence test requires the old host in SKILL.md, so fixing the links breaks the build.…
Read the thread · 2026-09-24 · open · outside contributor · 1 comment
feat(cli,mcp): heliograph cancel [id] and heliograph stop
There is no CLI verb or MCP tool to cancel a running step or stop the station loop. SKILL.md fills the gap by telling the agent to hand-edit station/request, commit and push. That contradicts the skill's own rule against editing files instead of using the CLI, and commit-and-push means nothing on the share, bundle, object store or relay transports. An agent that needs to kill a runaway step on those transports has no way to do it.
Evidence
skills/heliograph/SKILL.md:144-149: "set…
Read the thread · 2026-09-24 · closed · outside contributor · 0 comments
fix: the flagship net-probe example is refused, for the wrong reason
The example every doc leads with, heliograph send net-probe, does not resolve on a stock station: run.sh registers the step as net. The station then hides the real cause. step_mode throws away the "unknown step" exit, so the refusal reads "step 'net-probe' declares no mode ()". status and watch never print the reason: field, so the user sees generic advice about --allow-actions and CONFIRM=yes, and status says "Its reason is above" when it is not.
Evidence -…
Read the thread · 2026-09-24 · closed · outside contributor · 0 comments
fix(mcp): heliograph_send requests never expire and carry no target
The MCP heliograph_send tool builds its request by hand and leaves out three fields the CLI sets: Target, Expires and Mode. It also skips Validate(). The CLI's own comment says why the first two matter: without them a request lifted out of a transport repo stays valid for ever, and a replayed destructive step runs with every gate already satisfied. An agent driving heliograph over MCP gets none of that protection.
Evidence
cmd/heliograph/mcptools.go:121-127:wire.Request…
Read the thread · 2026-09-24 · closed · outside contributor · 0 comments
A blocked port is diagnosed as a credential problem, and 443 is never mentioned
Both git schemes are already supported and diagnosed: tp_preflight branches on the remote (git@*|ssh://* against https://*), reports the ssh-agent state with four outcomes and remedies, and reports the token mechanism for https by kind and length. That half is done and this issue is not about adding it.
The gap is the network layer underneath it. An estate that blocks outbound 22 is ordinary, and it is exactly the estate this tool is for.
What the operator sees today.…
Read the thread · 2026-09-10 · closed · 0 comments
Most recent
The remaining reports are on the project's issue tracker.