get-task-context

id	retrieval/get-task-context
name	Get Task Context
description	Build a context pack for a coding agent task with cited decisions, lessons, patterns, and repo/project context.
status	active
parameters	[{"name":"task","type":"string","required":true},{"name":"repo","type":"string","required":false},{"name":"project","type":"string","required":false},{"name":"include","type":"string[]","required":false,"default":["decisions","lessons","patterns"]},{"name":"maxTokens","type":"number","required":false,"default":6000}]
outputs	[{"name":"contextPack","type":"ContextPack"}]
uses	["retrieval.getTaskContext"]
steps	[{"id":"load-index","kind":"deterministic","tool":"retrieval.loadOrBuildIndex"},{"id":"search","kind":"deterministic","tool":"retrieval.searchPages"},{"id":"render","kind":"deterministic","tool":"retrieval.getTaskContext"}]
lanes	{"covers":["task-context","hermes-context-primary"],"excludes":["repo-only-context","single-page-fetch","write-anything"]}
risk	low
governance	none
triggers	["get context for this repo task","build a context pack","what should I know before working on ..."]

Get Task Context

Purpose

Synthesize a Markdown context pack for a coding-agent task. Cites local file paths. Penalizes deprecated/superseded pages (already in ranking) and labels them when included.

When to Use

Before any non-trivial coding task — especially when calling an LLM that benefits from anchored context. This is the primary integration point with Hermes.

Inputs

• task — required, free text.
• repo, project — optional scope filters.
• include — bucket selector; defaults to decisions/lessons/patterns.
• maxTokens — default 6000; hard cap 12000.

Procedure

1. Load (or build) the JSON index for the vault.
2. For each include bucket, run a typed search filtered by repo/project.
3. Render Markdown with sections: Task → Context Summary → Repo/Project Context → buckets → Pitfalls → Files to Inspect → Open Questions.
4. Compute confidence: low if <2 active relevant pages, medium if 2, high if ≥3.
5. Truncate buckets if estimated tokens exceed budget.

Output Format

A ContextPack JSON object whose contextPackMarkdown field is what an agent should paste into its planning prompt.

Failure Modes

• No vault → fall back to empty pack with confidence: low.
• All matches deprecated → still include (with [deprecated] label) and warn.

Examples

get task context for "implement run cancellation" --repo hermes-agent → Markdown with Compiled-Truth excerpts and citations.

Related Skills

• retrieval/repo-context — when you only need the repo summary.
• integrations/hermes-context — wraps this skill for Hermes' planning prompt.