SREPD

A PagerDuty terminal user interface focused on common SRE tasks.

Features

• View and manage PagerDuty incidents with team and individual views
• Acknowledge, re-escalate, silence, and merge incidents with confirmation prompts
• Open SOP/runbook links and incidents directly from alerts
• Log into clusters via ocm-container or ocm backplane with multi-cluster selection
• Add notes, auto-refresh with selection preservation, auto-acknowledge when on-call
• Background data freshness: incident details, alerts, and notes are cached and automatically re-fetched when older than five minutes
• PagerDuty environment variables passed automatically to terminal sessions
• Flag conditions: mark incidents matching cluster ID or organization name patterns
• AI agents: :agent CLI queries and :watcher LLM analysis with ambient incident pattern detection
• OCM integration: cluster enrichment with display names, service logs, limited support history
• Backplane integration: CORA cluster diagnostic reports via backplane API
• 8-tab incident viewer: Details, Alerts, Notes, Cluster, SLs, LS History, Reports, PD History
• Auto-update notification and srepd update self-update command
• Full configuration reference

Installation

make install        # install to $GOPATH/bin
# or
go install .        # standard go install

Getting Started

1. Create a PagerDuty API User Token: PagerDuty web → My Profile → User Settings → API Access → Create New API User Token.
2. Run srepd — the setup wizard does the rest: it validates your token, discovers your teams and escalation policies, detects your terminal and editor, and drops you straight into the live incident view when you save.

That's it. Your team may also publish a preset with its policy decisions — if so, run srepd config --preset <url> from your team's onboarding docs instead. See docs/presets.md for details.

Commands

Command	Description
srepd	Start the TUI
srepd update	Update to the latest release in place
srepd --version	Print version and git SHA
srepd --dev	Run with fixture data (no PD connection)
srepd config	Interactive configuration wizard (--preset <file|https-url> to pre-seed from a team preset)
srepd config generate	Print a complete annotated config with defaults (--out <path> to write a file)

Configuration

SREPD reads ~/.config/srepd/srepd.yaml and supports SREPD_ environment variable prefix. Run srepd config to create or update your config interactively. Values from environment variables (e.g., SREPD_TOKEN) are pre-filled automatically.

If no config file exists — or the config file is incomplete or still contains placeholder values — running srepd automatically enters the configuration wizard. The wizard validates your token, greets you by name, auto-selects your team when you belong to exactly one, and gates advanced options (custom service-to-policy mappings) behind a default-No confirm so most users never see them. New config files are written using the same annotated template as srepd config generate — with section headers, detected terminal alternatives as comments, and environment-aware defaults. The form resizes dynamically when the terminal window changes size.

Migrating from old config format: If your config uses the deprecated service_escalation_policies key, running srepd config will automatically migrate to the new default_silent_escalation_policy and custom_service_escalation_policies keys. The old block is commented out with a deprecation note.

Team presets: srepd config --preset <file|https-url> pre-seeds the wizard from a team-published YAML fragment carrying team policy decisions — teams, default_silent_escalation_policy, custom_service_escalation_policies, cluster_login_command, and optionally terminal/editor. Presets can never set token or llm_api credentials, only fill values you haven't configured, and every value is still confirmed in the wizard. URL fetches are HTTPS-only and size-capped. Teams can publish one preset link in their onboarding docs to eliminate the ID scavenger hunt for new members.

Because terminal, editor, and cluster_login_command are commands srepd executes, a preset that seeds any of them triggers an extra safety gate after the final "Save changes?" confirmation: a bold red warning listing every preset-supplied command for review, followed by an explicit "Are you sure you trust the source?" confirmation showing the preset file or URL. Both default to No, and declining either discards all changes. Values you type yourself, and preset fields that are only PagerDuty IDs (teams, policies, mappings), never trigger the gate.

Required

Key	Type	Description
token	string	PagerDuty API OAuth token
teams	[]string	PagerDuty team IDs

Optional

Key	Type	Default	Description
default_silent_escalation_policy	string	(none)	Silent escalation policy ID for silencing incidents. The wizard fetches policies from all your PagerDuty teams and marks ones with no on-call schedules as candidates ("does not page"); you can also skip or enter an ID manually.
custom_service_escalation_policies	map[string]string	(none)	Per-service silent policy overrides (service ID to policy ID)
editor	string	vim	Editor for incident notes (wizard prefills from $EDITOR/$VISUAL)
terminal	string	gnome-terminal --	Terminal emulator for cluster login (wizard detects installed terminals and warns when the configured one is missing)
cluster_login_command	string	ocm backplane login %%CLUSTER_ID%%	Cluster login command
rosa_boundary_command	string	rosa-boundary start-task --cluster-id %%CLUSTER_ID%% --connect	rosa-boundary cluster login command (opens in a new terminal window, like cluster_login_command)
toolbox_mode	string	auto	Toolbox detection: auto, true, or false
chord_prefix	string	ctrl+x	Prefix key for chord commands
agent_cli_command	string	claude --print	CLI agent command for :agent queries (set to "" to disable AI features)
emoji	bool	true	Use emoji markers or text fallbacks for flags/agent/watcher
reescalate_level	int	2	Escalation level ctrl+e re-escalates to, skipping lower placeholder tiers (e.g. level 1 "Nobody")
stream_responses	bool	true	Stream :watcher and :agent responses token-by-token (:watcher requires a streaming-capable provider; :agent auto-detects Claude CLI); set false for blocking responses
agent_system_prompt	string	(read-only investigation)	System prompt for :agent CLI queries
watcher_system_prompt	string	(SRE assistant)	System prompt for :watcher LLM queries
colors	map[string]string	(defaults)	Custom color scheme (hex values)

See docs/configuration.md for the full reference including CLI arguments.

Colors

All color keys are optional. Unspecified keys use defaults. Values must be hex colors (e.g., #778da9). See docs/colors.md for color key descriptions and pre-built palettes (Nord, Catppuccin Latte).

colors:
  text: "#778da9"       # Normal text, table rows
  border: "#415a77"     # Borders, tab outlines, separators
  highlight: "#ffffff"  # Headers, selected row text, active tab
  selected: "#415a77"   # Selected row background
  warning: "#a4133c"    # Warning/confirmation prompts
  error: "#0d1b2a"      # Error modal background
  muted: "#5C5C5C"      # Muted text (version, incident ID)
  tab: "#7D56F4"        # Reserved for future use

Advanced: manual configuration

Prefer editing a file over the wizard? Generate a complete, annotated config with every supported key and its default:

srepd config generate --out ~/.config/srepd/srepd.yaml

Then fill in token and teams. Do not copy placeholder values like <PagerDuty API token> into the file — srepd treats them as unconfigured and routes you into the wizard. A minimal working config looks like:

token: u+yourRealTokenHere
teams:
  - PABC123
default_silent_escalation_policy: P654321
terminal: gnome-terminal --
cluster_login_command: ocm backplane login %%CLUSTER_ID%%
rosa_boundary_command: rosa-boundary start-task --cluster-id %%CLUSTER_ID%% --connect
toolbox_mode: auto

Use %%CLUSTER_ID%% and %%INCIDENT_ID%% as placeholders in terminal or cluster_login_command for dynamic substitution at launch time. If %%CLUSTER_ID%% is not present in cluster_login_command, the cluster ID is appended automatically.

Terminal Support

Terminal profiles are auto-detected from the executable name or Flatpak app ID. Set only the terminal name in your config; the correct argument style is handled automatically.

Supported terminals: gnome-terminal, ptyxis, wezterm, blackbox, tmux, konsole, alacritty, ghostty, terminator, kitty, foot, contour, iterm2, macOS Terminal

Flatpak-installed terminals are also supported using their application ID (e.g., org.kde.konsole).

When running inside a Fedora Toolbox, terminal commands are automatically prefixed with flatpak-spawn --host (controlled by toolbox_mode).

OCM Integration

SREPD enriches PagerDuty incident data with cluster details from the OpenShift Cluster Manager (OCM) API. On startup, it connects to the production OCM API using tokens from ~/.config/ocm/ocm.json. If tokens are expired, a browser window opens for auth code login in the background — the TUI starts immediately with PagerDuty incidents visible, and OCM enrichment populates once authentication completes. OCM authentication is skipped while the configuration wizard is on screen (srepd config and first-run setup) so new users aren't interrupted before saving a token; it runs automatically as soon as the wizard completes and the live session begins.

Enriched data includes:

• Cluster display names replace PD service names in the incident table (e.g., mycluster.abc1.p1.example.org instead of osd-mycluster.abc1.p1.example.org-hive-cluster)
• Impacted Clusters section on the Details tab lists all clusters in the incident
• Cluster tab shows OCM cluster details: name, ID, state, region, provider, version, CCS, Hypershift
• Service Logs tab shows recent service logs per cluster
• Limited Support History tab shows LS reasons per cluster
• Reports tab shows CORA cluster diagnostic reports from the backplane API
• Multi-cluster incidents show (+N) in the service column

OCM features are optional — if OCM is not configured, the remaining TUI functions normally.

The Reports tab requires ~/.config/backplane/config.json (standard backplane-cli config). The backplane URL is resolved from OCM environment metadata. If the config file is missing, the tab shows a message indicating backplane is not enabled.

LLM Integration

SREPD supports configurable LLM providers for AI-assisted incident analysis and an ambient watcher that detects cross-incident patterns. Configuration is entirely optional — AI features are disabled when unconfigured. See docs/ai-agents.md for usage and docs/llm-providers.md for provider setup.

Use :agent <query> for CLI agent queries and :watcher <query> for LLM analysis. Press w to toggle the watcher pane.

Quick Start

Add to ~/.config/srepd/srepd.yaml:

llm_api:
  provider: ollama
  model: llama3.1:8b

Supported Providers

Provider	Endpoint Default	Description
ollama	http://localhost:11434	Local Ollama daemon (or ollama-container)
anthropic	SDK default	Anthropic Messages API (direct API key)
anthropic-vertex	auto	Anthropic via Google Cloud Vertex AI (uses Application Default Credentials)
anthropic-bedrock	auto	Anthropic via AWS Bedrock (uses AWS default credential chain)
openai	(user must specify)	Any OpenAI-compatible endpoint (vLLM, TGI, internal models)
ramalama	http://localhost:8080	Red Hat RamaLama server

Configuration Reference

Key	Type	Description
llm_api.provider	string	Provider name: ollama, anthropic, anthropic-vertex, anthropic-bedrock, openai, ramalama
llm_api.endpoint	string	API endpoint URL (provider-specific default used if omitted)
llm_api.model	string	Model identifier (provider-specific default used if omitted)
llm_api.api_key_env	string	Name of env var containing API key (optional for all providers)
llm_api.region	string	Cloud region for anthropic-vertex (auto-detected from CLOUD_ML_REGION or VERTEXAI_LOCATION). Not used for anthropic-bedrock — its region comes from the AWS SDK environment
llm_api.project_id	string	GCP project for anthropic-vertex (auto-detected from ANTHROPIC_VERTEX_PROJECT_ID, VERTEXAI_PROJECT, or GOOGLE_CLOUD_PROJECT)

AWS Bedrock: export your AWS credentials and region in the environment (AWS_PROFILE / AWS_REGION) — none go in the config. Or authenticate with a Bedrock API key by exporting AWS_BEARER_TOKEN_BEDROCK (plus AWS_REGION) instead of IAM credentials. Enable Anthropic models once in the AWS Bedrock console, and set model to an inference-profile ID such as us.anthropic.claude-sonnet-4-6 (bare foundation-model IDs are not on-demand invocable). See LLM Providers for full setup.

Key Bindings

Press h to toggle the help overlay inside srepd.

Key	Action	Key	Action
j/k	Move down/up	a	Acknowledge
g/G	Jump to top/bottom	n	Add note
Enter	View incident	l	Login to cluster
Esc	Go back	o	Open in browser
h	Toggle help	s	Open SOP link
t	Toggle team/individual	ctrl+s	Silence
r	Refresh	ctrl+e	Re-escalate
ctrl+r	Toggle auto-refresh	u	Toggle urgency filter
ctrl+a	Toggle auto-acknowledge	ctrl+l	View debug log
ctrl+q/ctrl+c	Quit	1-9	Select cluster
://	Command input	m	Merge incident
w	Toggle watcher pane	ctrl+t	Add tags to incident
ctrl+x + key	Chord commands	ctrl+x ?	Show chord help
Tab/Shift+Tab/←/→	Switch tabs (incident view)	↑/↓	Scroll within tab

Chord commands use a configurable prefix (default ctrl+x) followed by a second key. Set chord_prefix in config to change.

rosa-boundary Support

SREPD supports rosa-boundary as an alternative cluster login method. Use ctrl+x b to open a rosa-boundary session for the selected incident's cluster. The default command is rosa-boundary start-task --cluster-id %%CLUSTER_ID%% --connect. Override via the rosa_boundary_command config key or SREPD_ROSA_BOUNDARY_COMMAND environment variable.

rosa-boundary launches an interactive session into a protected cluster exactly like ocm-container, and behaves identically to cluster_login_command: the session opens in a new terminal window, SREPD keeps running, and multiple concurrent sessions are supported. The same PagerDuty context passed to ocm-container sessions (the PAGERDUTY_* environment variables below) is passed to rosa-boundary sessions; unsupported variables are simply ignored. In toolbox mode the terminal launches on the host via flatpak-spawn --host, the same convention as the cluster login path.

Environment Variables

When using ocm-container, PagerDuty context is passed automatically as environment variables:

Variable	Description
PAGERDUTY_INCIDENT_ID	Incident ID
PAGERDUTY_INCIDENT_TITLE	Incident title (sanitized)
PAGERDUTY_INCIDENT_URL	Direct link to the incident
PAGERDUTY_INCIDENT_SERVICE	Service name
PAGERDUTY_INCIDENT_URGENCY	high or low
PAGERDUTY_INCIDENT_STATUS	triggered, acknowledged, or resolved
PAGERDUTY_CLUSTER_ID	Selected cluster ID
PAGERDUTY_ALERT_COUNT	Number of matching alerts
PAGERDUTY_ALERT_NAMES	Comma-separated alert names
PAGERDUTY_ALERT_LINKS	Comma-separated SOP/runbook links
PAGERDUTY_NOTES_EXIST	true or false
PAGERDUTY_NOTE_COUNT	Number of notes
PAGERDUTY_CLAUDE_AVAILABLE	true when Claude Code is detected on PATH
REASON	Incident URL (for compliance/audit)

For non-ocm-container terminals, environment variables are set on the process directly or passed via flatpak-spawn --env= when in toolbox mode.

Development

Command	Purpose
make build	Build via goreleaser snapshot
make install	Install to $GOPATH/bin
make test	Run unit tests
make lint	Run golangci-lint
make vet	Run go vet
make fmt	Format code
make fmt-check	Check formatting (CI-friendly)
make coverage	Generate coverage report
make test-all	Run all checks: fmt-check, vet, lint, test
make plan-check	Verify plan document exists for branch
make clean	Remove build artifacts
make help	Show all available targets

PRs follow TDD workflow and require a plan document in docs/plans/. Run make test-all before pushing.

License

MIT License. See LICENSE for details.

A Note on AI Contributions

Portions of this codebase were developed with the assistance of AI tools, including Claude by Anthropic. To the extent that AI-generated contributions are subject to copyright and capable of being licensed under applicable law, such contributions are licensed under the same MIT License terms stated above. The legal status of AI-generated code remains an evolving area of law, and this notice is provided in the interest of transparency. All AI-assisted contributions have been reviewed and approved by the project maintainers.