Add context intelligence: symbol-aware packets, context diff and score

[00PrabalK00]

Why

Strategy §Gap3 / §Phase2: make the context an agent receives an inspectable
object, not a hidden prompt. This adds three capabilities on top of the
existing MemoryStore and context build packet: smarter (symbol-aware)
packets, packet diffing, and packet scoring.

Constraints honored: stdlib + git subprocess only (no tree-sitter, no
ctags, no new deps). Symbol extraction is conservative regex over file text;
token estimate reuses the repo's chars // 4 heuristic (core.estimate_tokens,
tied to CONTEXT_BUDGETS).

What

New module continuum/context_intel.py (pure functions): extract_symbols,
relevant_tests, recent_commits, gather_context_intel, diff_intel,
score_intel plus renderers. Three CLI commands under continuum context:

• continuum context enrich <task_ref> — relevant files (claimed + changed),
  extracted symbols, linked tests, recent commits touching those files, prior
  decisions (DECISIONS.md + decision events), current blockers. Human + --json.
  Accepts --file to analyze explicit paths instead of a task's claims.
• continuum context diff <ref_a> <ref_b> — files/symbols/tests/decisions
  only-in-A vs only-in-B and token-size delta. Answers "who got stale context".
  Human + --json.
• continuum context score <ref> — estimated_tokens, source_count,
  staleness (hours/days), deterministic risk_level (low/med/high from
  out-of-scope files, missing tests, stale context, large size) and a
  missing_info checklist. Human + --json. Non-zero exit only on real error
  (unknown task).
• continuum context build --intel/--symbols — appends a compact
  "Relevant Symbols / Tests / Recent Changes" section. Default packet behavior
  is unchanged.

Real smoke output

context enrich T0001 (task claiming auth.py):

# Context Intel: T0001 Auth work
Status: RUNNING  Agent: claude
## Relevant Symbols
- `auth.py`: def login, class AuthService, const MAX_RETRIES
## Linked Tests
- `tests/test_auth.py`
## Recent Changes
- 4c71ae5 add auth and test

context diff T0001 T0002:

## Symbols
Only in T0001:
  - auth.py::class AuthService
  - auth.py::const MAX_RETRIES
  - auth.py::def login
Only in T0002:
  - ui.js::const HANDLER
  - ui.js::function render
## Token Size
- T0001: 87 tokens
- T0002: 81 tokens
- Delta (A - B): 6 tokens

context score T0002 --json (task with no linked tests):

{
  "estimated_tokens": 81,
  "source_count": 3,
  "sources": ["files", "symbols", "recent_commits"],
  "staleness_hours": 0.01,
  "risk_level": "low",
  "risk_reasons": ["no linked tests"],
  "missing_info": ["no tests linked", "no recent decisions"]
}

Unknown task exits non-zero: context score T9999 -> Error: Unknown task: T9999, exit 1.

Tests

tests/test_context_intel.py adds 17 tests (symbol extraction for Python + JS
ignoring comments/strings; test<->module linking; recent_commits in a git repo
and [] in a non-git dir; enrich/diff/score CLI field + JSON shape + exit codes;
score flags missing_info when sections are empty). Git-based tests set repo
identity in setUp (CI has no global identity).

CI-mimic GIT_CONFIG_GLOBAL=/dev/null GIT_CONFIG_SYSTEM=/dev/null python -m unittest discover -s tests: 246 tests pass (229 prior + 17 new).

🤖 Generated with Claude Code